Load the required libraries.
library(tidyverse)
library(sf)
library(here)
library(readxl)
library(scales)
library(DT)
library(brms)
library(tidybayes)
library(patchwork)
library(marginaleffects)
library(ggrepel)
library(scico)
library(ggdensity)
library(ggpubr)
library(units)
#library(ggsn)
Functions that we will use throughout the script
#labeller for years
year_labels <- c(1950:1963)
#The Glasgow mass minuture chest X-ray campaign happened between 11th March and 12th April 1957
#Segment for graphs to match ACF period
acf_start <- decimal_date(ymd("1957-03-11"))
acf_end <- decimal_date(ymd("1957-04-12"))
Function for counterfactual plots
plot_counterfactual <- function(model_data, model, population_denominator, outcome, grouping_var=NULL, re_formula,...){
#labeller for years
year_labels <- c(1950:1963)
#The Glasgow mass minuture chest X-ray campaign happened between 11th March and 12th April 1957
#Segment for graphs to match ACF period
acf_start <- decimal_date(ymd("1957-03-11"))
acf_end <- decimal_date(ymd("1957-04-12"))
summary <- {{model_data}} %>%
select(year, year2, y_num, acf_period, {{population_denominator}}, {{outcome}}, {{grouping_var}}) %>%
add_epred_draws({{model}}, re_formula={{re_formula}}) %>%
group_by(year2, acf_period, {{grouping_var}}) %>%
mean_qi() %>%
mutate(.epred_inc = .epred/{{population_denominator}}*100000,
.epred_inc.lower = .epred.lower/{{population_denominator}}*100000,
.epred_inc.upper = .epred.upper/{{population_denominator}}*100000) %>%
mutate(acf_period = case_when(acf_period=="a. pre-acf" ~ "Before Intervention",
acf_period=="c. post-acf" ~ "Post Intervention"))
#create the counterfactual (no intervention), and summarise
counterfact <-
add_epred_draws(object = {{model}},
newdata = {{model_data}} %>%
select(year, year2, y_num, {{population_denominator}}, {{grouping_var}}, {{outcome}}) %>%
mutate(acf_period = "a. pre-acf"), re_formula={{re_formula}}) %>%
group_by(year2, acf_period, {{grouping_var}}) %>%
mean_qi() %>%
mutate(.epred_inc = .epred/{{population_denominator}}*100000,
.epred_inc.lower = .epred.lower/{{population_denominator}}*100000,
.epred_inc.upper = .epred.upper/{{population_denominator}}*100000) %>%
mutate(acf_period = case_when(acf_period=="a. pre-acf" ~ "Before Intervention",
acf_period=="c. post-acf" ~ "Post Intervention"))
#plot the intervention effect
p <- summary %>%
droplevels() %>%
ggplot() +
geom_ribbon(aes(ymin=.epred_inc.lower, ymax=.epred_inc.upper, x=year2, group = acf_period, fill=acf_period), alpha=0.5) +
geom_ribbon(data = counterfact %>% filter(year>=1956),
aes(ymin=.epred_inc.lower, ymax=.epred_inc.upper, x=year2, fill="Counterfactual"), alpha=0.5) +
geom_line(data = counterfact %>% filter(year>=1956),
aes(y=.epred_inc, x=year2, colour="Counterfactual")) +
geom_line(aes(y=.epred_inc, x=year2, group=acf_period, colour=acf_period)) +
geom_point(data = {{model_data}}, aes(y={{outcome}}, x=year2, shape=acf_period), size=2) +
geom_vline(aes(xintercept=acf_end), linetype=3) +
theme_ggdist() +
scale_y_continuous(labels=comma, limits = c(0,NA)) +
scale_x_continuous(labels = year_labels,
breaks = year_labels) +
scale_fill_manual(values = c("#DE0D92", "grey50", "#4D6CFA") , name="", na.translate = F) +
scale_colour_manual(values = c("#DE0D92", "grey50", "#4D6CFA") , name="", na.translate = F) +
scale_shape_discrete(name="", na.translate = F) +
labs(
x = "Year",
y = "Case notification rate (per 100,000)",
caption = "Mass miniature X-ray campaign period between dashed lines (11th March-12th April 1957)"
) +
theme(legend.position = "bottom",
panel.border = element_rect(colour = "grey78", fill=NA),
text = element_text(size=8),
axis.text.x = element_text(size=8, angle = 90, hjust=1, vjust=0.5),
legend.text = element_text(size=8)) +
guides(shape="none")
facet_vars <- vars(...)
if (length(facet_vars) != 0) {
p <- p + facet_wrap(facet_vars)
}
p
}
Function for calculating measures of change over time (RR.peak, RR.level, RR.slope)
summarise_change <- function(model_data, model, population_denominator, grouping_var = NULL, re_formula = NULL) {
#functions for calculating RR.peak
#i.e. relative case notification rate in 1957 vs. counterfactual trend for 1957
grouping_var <- enquo(grouping_var)
if (!is.null({{grouping_var}})) {
#make the prediction matrix, conditional on whether we want random effects included or not.
out <- crossing({{model_data}} %>%
select({{population_denominator}}, y_num, !!grouping_var) %>%
filter(y_num == 8),
acf_period = c("a. pre-acf", "b. acf")
)
} else {
out <- crossing({{model_data}} %>%
select({{population_denominator}}, y_num) %>%
filter(y_num == 8),
acf_period = c("a. pre-acf", "b. acf")
)
}
peak_draws <- add_epred_draws(newdata = out,
object = {{model}},
re_formula = {{re_formula}}) %>%
mutate(epred_cnr = .epred/population_without_inst_ship*100000) %>%
group_by(.draw, !!grouping_var) %>%
summarise(estimate = last(epred_cnr)/first(epred_cnr)) %>%
ungroup() %>%
mutate(measure = "RR.peak")
peak_summary <- peak_draws %>%
group_by(!!grouping_var) %>%
mean_qi(estimate) %>%
mutate(measure = "RR.peak")
#functions for calculating RR.step
#i.e. relative case notification rate in 1958 vs. counterfactual trend for 1958
if (!is.null({{grouping_var}})) {
out2 <- crossing({{model_data}} %>%
select({{population_denominator}}, y_num, !!grouping_var) %>%
filter(y_num == 9),
acf_period = c("a. pre-acf", "c. post-acf")
)
} else {
out2 <- crossing({{model_data}} %>%
select({{population_denominator}}, y_num) %>%
filter(y_num == 9),
acf_period = c("a. pre-acf", "c. post-acf")
)
}
level_draws <- add_epred_draws(newdata = out2,
object = {{model}},
re_formula = {{re_formula}}) %>%
arrange(y_num, .draw) %>%
mutate(epred_cnr = .epred/population_without_inst_ship*100000) %>%
group_by(.draw, !!grouping_var) %>%
summarise(estimate = last(epred_cnr)/first(epred_cnr)) %>%
ungroup() %>%
mutate(measure = "RR.level")
level_summary <- level_draws %>%
group_by(!!grouping_var) %>%
mean_qi(estimate) %>%
mutate(measure = "RR.level")
#functions for calculating RR.slope
#i.e. relative change in case notification rate in 1958-1963 vs. counterfactual trend for 1959-1963
if (!is.null({{grouping_var}})) {
out3 <- crossing({{model_data}} %>%
select({{population_denominator}}, y_num, !!grouping_var) %>%
filter(y_num %in% c(9,14)),
acf_period = c("a. pre-acf", "c. post-acf")
)
} else {
out3 <- crossing({{model_data}} %>%
select({{population_denominator}}, y_num) %>%
filter(y_num %in% c(9,14)),
acf_period = c("a. pre-acf", "c. post-acf")
)
}
slope_draws <- add_epred_draws(newdata = out3,
object = {{model}},
re_formula = {{re_formula}}) %>%
arrange(y_num) %>%
ungroup() %>%
mutate(epred_cnr = .epred/population_without_inst_ship*100000) %>%
group_by(.draw, acf_period, !!grouping_var) %>%
summarise(slope = (last(epred_cnr) - first(epred_cnr)) / (last(y_num)-first(y_num))) %>%
ungroup() %>%
group_by(.draw, !!grouping_var) %>%
summarise(estimate = last(slope)/first(slope)) %>%
mutate(measure = "RR.slope")
slope_summary <- slope_draws %>%
group_by(!!grouping_var) %>%
median_qi(estimate) %>%
mutate(measure = "RR.slope")
#gather all the results into a named list
lst(peak_draws=peak_draws, peak_summary=peak_summary,
level_draws=level_draws, level_summary=level_summary,
slope_draws=slope_draws, slope_summary=slope_summary)
}
Function for calculating difference from counterfactual
calculate_counterfactual <- function(model_data, model, population_denominator, grouping_var=NULL, re_formula=NA){
#effect vs. counterfactual
counterfact <-
add_epred_draws(object = {{model}},
newdata = {{model_data}} %>%
select(year, year2, y_num, {{population_denominator}}, {{grouping_var}}) %>%
mutate(acf_period = "a. pre-acf"),
re_formula = {{re_formula}}) %>%
group_by(.draw, year, {{grouping_var}}, acf_period) %>%
mutate(.epred_inc_counterf = .epred/{{population_denominator}}*100000, .epred_counterf=.epred) %>%
filter(year>1957) %>%
ungroup() %>%
select(year, {{population_denominator}}, .draw, .epred_counterf, .epred_inc_counterf, {{grouping_var}})
#Calcuate case notification rate per draw, then summarise.
post_change <-
add_epred_draws(object = {{model}},
newdata = {{model_data}} %>%
select(year, year2, y_num, {{population_denominator}}, {{grouping_var}}, acf_period),
re_formula = {{re_formula}}) %>%
group_by(.draw, year, {{grouping_var}}, acf_period) %>%
mutate(.epred_inc = .epred/{{population_denominator}}*100000) %>%
filter(year>1957) %>%
ungroup() %>%
select(year, {{population_denominator}}, {{grouping_var}}, .draw, .epred, .epred_inc, {{grouping_var}})
#for the overall period
counterfact_overall <-
add_epred_draws(object = {{model}},
newdata = {{model_data}} %>%
select(year, year2, y_num, {{population_denominator}}, {{grouping_var}}) %>%
mutate(acf_period = "a. pre-acf"),
re_formula = {{re_formula}}) %>%
group_by(.draw, {{grouping_var}}) %>%
filter(year>1957) %>%
ungroup() %>%
select({{population_denominator}}, .draw, .epred, {{grouping_var}}) %>%
group_by(.draw, {{grouping_var}}) %>%
summarise(.epred_counterf = sum(.epred))
#Calcuate case notification rate per draw, then summarise.
post_change_overall <-
add_epred_draws(object = {{model}},
newdata = {{model_data}} %>%
select(year, year2, y_num, {{population_denominator}}, {{grouping_var}}, acf_period),
re_formula = {{re_formula}}) %>%
group_by(.draw, {{grouping_var}}) %>%
filter(year>1957) %>%
ungroup() %>%
select({{population_denominator}}, {{grouping_var}}, .draw, .epred) %>%
group_by(.draw, {{grouping_var}}) %>%
summarise(.epred = sum(.epred))
counter_post <-
left_join(counterfact, post_change) %>%
mutate(cases_averted = .epred_counterf-.epred,
pct_change = (.epred - .epred_counterf)/.epred_counterf,
diff_inc100k = .epred_inc - .epred_inc_counterf,
rr_inc100k = .epred_inc/.epred_inc_counterf) %>%
group_by(year, {{grouping_var}}) %>%
mean_qi(cases_averted, pct_change, diff_inc100k, rr_inc100k) %>%
ungroup()
counter_post_overall <-
left_join(counterfact_overall, post_change_overall) %>%
mutate(cases_averted = .epred_counterf-.epred,
pct_change = (.epred - .epred_counterf)/.epred_counterf) %>%
group_by({{grouping_var}}) %>%
mean_qi(cases_averted, pct_change) %>%
ungroup()
lst(counter_post, counter_post_overall)
}
Function for tidying up counterfactuals (mostly for making nice tables)
tidy_counterfactuals <- function(data){
data %>%
mutate(across(c(cases_averted:cases_averted.upper, diff_inc100k:diff_inc100k.upper), number_format(accuracy = 0.1, big.mark = ","))) %>%
mutate(across(c(rr_inc100k:rr_inc100k.upper), number_format(accuracy = 0.01))) %>%
mutate(across(c(pct_change:pct_change.upper), percent, accuracy=0.1)) %>%
mutate(year = as.character(year),
cases_averted = glue::glue("{cases_averted} ({cases_averted.lower} to {cases_averted.upper})"),
pct_change = glue::glue("{pct_change} ({pct_change.lower} to {pct_change.upper})"),
diff_inc = glue::glue("{diff_inc100k} ({diff_inc100k.lower} to {diff_inc100k.upper})"),
rr_inc = glue::glue("{rr_inc100k} ({rr_inc100k.lower} to {rr_inc100k.upper})"))
}
tidy_counterfactuals_overall <- function(data){
data %>%
mutate(across(c(cases_averted:cases_averted.upper), number_format(accuracy = 0.1, big.mark = ","))) %>%
mutate(across(c(pct_change:pct_change.upper), percent, accuracy=0.1)) %>%
mutate(year = as.character(year),
cases_averted = glue::glue("{cases_averted} ({cases_averted.lower} to {cases_averted.upper})"),
pct_change = glue::glue("{pct_change} ({pct_change.lower} to {pct_change.upper})"))
}
Import datasets for analysis
Make a map of Glasgow wards
glasgow_wards_1951 <- st_read(here("mapping/glasgow_wards_1951.geojson"))
Reading layer `glasgow_wards_1951' from data source
`/Users/petermacpherson/Dropbox/Projects/Historical TB ACF 2023-11-28/Work/analysis/glasgow-cxr/mapping/glasgow_wards_1951.geojson'
using driver `GeoJSON'
Simple feature collection with 37 features and 3 fields
Geometry type: MULTIPOLYGON
Dimension: XY
Bounding box: xmin: -4.393502 ymin: 55.77464 xmax: -4.070411 ymax: 55.92814
Geodetic CRS: WGS 84
#read in Scotland boundary
scotland <- st_read(here("mapping/Scotland_boundary/Scotland boundary.shp"))
Reading layer `Scotland boundary' from data source
`/Users/petermacpherson/Dropbox/Projects/Historical TB ACF 2023-11-28/Work/analysis/glasgow-cxr/mapping/Scotland_boundary/Scotland boundary.shp'
using driver `ESRI Shapefile'
Simple feature collection with 1 feature and 1 field
Geometry type: MULTIPOLYGON
Dimension: XY
Bounding box: xmin: 5513 ymin: 530249 xmax: 470332 ymax: 1220302
Projected CRS: OSGB36 / British National Grid
#make a bounding box for Glasgow
bbox <- st_bbox(glasgow_wards_1951) |> st_as_sfc()
#plot scotland with a bounding box around the City of Glasgow
scotland_with_bbox <- ggplot() +
geom_sf(data = scotland, fill="antiquewhite") +
geom_sf(data = bbox, colour = "#C60C30", fill="antiquewhite") +
theme_void() +
theme(panel.border = element_rect(colour = "grey78", fill=NA, linewidth = 0.5),
panel.background = element_rect(fill = "#EAF7FA", size = 0.3))
Warning: The `size` argument of `element_rect()` is deprecated as of ggplot2 3.4.0.
Please use the `linewidth` argument instead.
#plot the wards
#note we tidy up some names to fit on map
glasgow_ward_map <- glasgow_wards_1951 %>%
mutate(ward = case_when(ward=="Shettleston and Tollcross" ~ "Shettleston and\nTollcross",
ward=="Partick (West)" ~ "Partick\n(West)",
ward=="Partick (East)" ~ "Partick\n(East)",
ward=="North Kelvin" ~ "North\nKelvin",
ward=="Kinning Park" ~ "Kinning\nPark",
TRUE ~ ward)) %>%
ggplot() +
geom_sf(aes(fill=division)) +
geom_sf_label(aes(label = ward), size=3, fill=NA, label.size = NA, colour="black") +
#scale_colour_identity() +
scale_fill_brewer(palette = "Set3", name="City of Glasgow Division") +
theme_grey() +
labs(x="",
y="",
fill="Division") +
theme(legend.position = "top",
panel.border = element_rect(colour = "grey78", fill=NA, linewidth = 0.5),
panel.background = element_rect(fill = "antiquewhite", size = 0.3),
panel.grid.major = element_line(color = "grey78")) +
guides(fill=guide_legend(title.position = "top", title.hjust = 0.5, title.theme = element_text(face="bold")))
#add the map of scotland as an inset
glasgow_ward_map + inset_element(scotland_with_bbox, 0.75, 0, 1, 0.4)
ggsave(here("figures/s1.png"), height=10, width = 12)
NA
NA
Calculate areas per geographical unit
sf_use_s2(FALSE) #https://github.com/r-spatial/sf/issues/1762
Spherical geometry (s2) switched off
glasgow_wards_1951 <- glasgow_wards_1951 %>%
mutate(area = st_area(glasgow_wards_1951))
glasgow_wards_1951$area_km <- units::set_units(glasgow_wards_1951$area, km^2)
Make division shape files, and calculate area (stopped working, need to fix!)
# glasgow_divisions_1951 <- glasgow_wards_1951 %>%
# group_by(division) %>%
# summarize(geometry = st_union(geometry)) %>%
# nngeo::st_remove_holes() %>%
# mutate(area = st_area(glasgow_divisions_1951))
#
# glasgow_divisions_1951$area_km <- units::set_units(glasgow_divisions_1951$area, km^2)
Load in the datasets for denonomiators, and check for consistency.
overall_pops <- read_xlsx(path = "2023-11-28_glasgow-acf.xlsx", sheet = "overall_population")
overall_pops %>%
mutate(across(where(is.numeric) & !(year), ~comma(.))) %>%
datatable()
#shift year to midpoint
overall_pops <- overall_pops %>%
mutate(year2 = year+0.5)
Note, we have three population estimates:
(Population in shipping is estimated from the 1951 census, so is the same for most years)
First, plot the total population
overall_pops %>%
ggplot() +
geom_area(aes(y=total_population, x=year2), alpha=0.5, colour = "mediumseagreen", fill="mediumseagreen") +
geom_point(aes(y=total_population, x=year2), colour = "mediumseagreen") +
geom_vline(aes(xintercept=acf_start), linetype=3) +
geom_vline(aes(xintercept=acf_end), linetype=3) +
scale_y_continuous(labels=comma) +
scale_x_continuous(labels = year_labels,
breaks = year_labels) +
labs(
title = "Glasgow Corporation: total population",
subtitle = "1950 to 1963",
x = "Year",
y = "Population",
caption = "Mid-year estimates\nMass miniature X-ray campaign period between dashed lines (11th March-12th April 1957)"
) +
theme_ggdist()
NA
NA
Now the population excluding institutionalised and shipping population
overall_pops %>%
ggplot() +
geom_area(aes(y=population_without_inst_ship, x=year2), alpha=0.5, colour = "purple", fill="purple") +
geom_point(aes(y=population_without_inst_ship, x=year2), colour = "purple") +
geom_vline(aes(xintercept=acf_start), linetype=3) +
geom_vline(aes(xintercept=acf_end), linetype=3) +
scale_y_continuous(labels=comma) +
scale_x_continuous(labels = year_labels,
breaks = year_labels) +
labs(
title = "Glasgow Corporation: population excluding institutionalised and shipping",
subtitle = "1950 to 1963",
x = "Year",
y = "Population",
caption = "Mid-year estimates\nMass miniature X-ray campaign period between dashed lines (11th March-12th April 1957)"
) +
theme_ggdist()
NA
NA
There are 5 Divisions containing 37 Wards in the Glasgow Corporation, with consistent boundaries over time.
#look-up table for divisions and wards
ward_lookup <- read_xlsx(path = "2023-11-28_glasgow-acf.xlsx", sheet = "divisions_wards")
ward_pops <- read_xlsx(path = "2023-11-28_glasgow-acf.xlsx", sheet = "ward_population")
ward_pops %>%
mutate(across(where(is.numeric) & !(year), ~comma(.))) %>%
datatable()
#shift year to midpoint
ward_pops <- ward_pops %>%
mutate(year2 = year+0.5)
#Get the Division population
division_pops <- ward_pops %>%
group_by(division, year) %>%
summarise(population_without_inst_ship = sum(population_without_inst_ship, na.rm = TRUE),
institutions = sum(institutions, na.rm = TRUE),
shipping = sum(shipping, na.rm = TRUE),
total_population = sum(total_population, na.rm = TRUE))
`summarise()` has grouped output by 'division'. You can override using the `.groups` argument.
division_pops %>%
mutate(across(where(is.numeric) & !(year), ~comma(.))) %>%
datatable()
NA
Plot the overall population by Division and Ward
division_pops %>%
mutate(year2 = year+0.5) %>%
ggplot() +
geom_area(aes(y=total_population, x=year2, colour=division, fill=division), alpha=0.8) +
geom_point(aes(y=total_population, x=year2, colour=division)) +
geom_vline(aes(xintercept=acf_start), linetype=3) +
geom_vline(aes(xintercept=acf_end), linetype=3) +
scale_y_continuous(labels=comma) +
facet_wrap(division~.) +
scale_x_continuous(labels = year_labels,
breaks = year_labels,
guide = guide_axis(angle = 90)) +
scale_fill_brewer(palette = "Set3", name = "") +
scale_colour_brewer(palette = "Set3", name = "") +
labs(
title = "Glasgow Corporation: total population by Division",
subtitle = "1950 to 1963",
x = "Year",
y = "Population",
caption = "Mid-year estimates\nMass miniature X-ray campaign period between dashed lines (11th March-12th April 1957)"
) +
theme_ggdist() +
theme(legend.position = "bottom")
NA
NA
ward_pops %>%
ggplot() +
geom_area(aes(y=total_population, x=year2, colour=division, fill=division), alpha=0.8) +
geom_point(aes(y=total_population, x=year2, colour=division)) +
geom_vline(aes(xintercept=acf_start), linetype=3) +
geom_vline(aes(xintercept=acf_end), linetype=3) +
scale_y_continuous(labels=comma) +
facet_wrap(ward~., ncol=6) +
scale_x_continuous(labels = year_labels,
breaks = year_labels,
guide = guide_axis(angle = 90)) +
scale_fill_brewer(palette = "Set3", name="Division") +
scale_colour_brewer(palette = "Set3", name = "Division") +
labs(
title = "Glasgow City: total population by Ward",
subtitle = "1950 to 1963",
x = "Year",
y = "Population",
caption = "Mid-year estimates\nMass miniature X-ray campaign period between dashed lines (11th March-12th April 1957)"
) +
theme_ggdist() +
theme(legend.position = "bottom")
ggsave(here("figures/s2.png"), height=10, width=12)
Approximately, how many person-years of follow-up do we have?
overall_pops %>%
ungroup() %>%
summarise(across(year, length, .names = "years"),
across(c(population_without_inst_ship, total_population), sum)) %>%
mutate(across(where(is.double), comma)) %>%
datatable()
NA
NA
Change in population by ward
ward_pops %>%
group_by(ward) %>%
summarise(pct_change_pop = (last(population_without_inst_ship) - first(population_without_inst_ship))/first(population_without_inst_ship)) %>%
mutate(pct_change_pop = percent(pct_change_pop)) %>%
arrange(pct_change_pop) %>%
datatable()
NA
NA
NA
Output population density by ward and divison for regression modelling
Wards first
(stopped working, need to fix)
# ward_covariates <- glasgow_wards_1951 %>%
# left_join(ward_pops) %>%
# mutate(people_per_km_sq = as.double(population_without_inst_ship/area_km))
#
# #plot it out
#
# ward_covariates %>%
# ggplot() +
# geom_sf(aes(fill=people_per_km_sq)) +
# facet_wrap(year~., ncol=7) +
# scale_fill_viridis_c(option="A") +
# theme(legend.position = "bottom",
# axis.text.x = element_text(angle = 45, hjust=1))
#
# ggsave(here("figures/ward_pop_density.png"), width=10)
#
# write_rds(ward_covariates, here("populations/ward_covariates.rds"))
Now divisions first
(stopped working, need to fix)
# division_covariates <- glasgow_divisions_1951 %>%
# left_join(division_pops) %>%
# mutate(people_per_km_sq = as.double(total_population/area_km))
#
# #plot it out
#
# division_covariates %>%
# ggplot() +
# geom_sf(aes(fill=people_per_km_sq)) +
# geom_sf_label(aes(label = division), size=3, fill=NA, label.size = NA, colour="black", family = "Segoe UI") +
# facet_wrap(year~., ncol=7) +
# scale_fill_viridis_c(option="G") +
# theme(legend.position = "bottom",
# axis.text.x = element_text(angle = 45, hjust=1))
#
# ggsave(here("figures/division_pop_density.png"), width=10)
#
# write_rds(division_covariates, here("populations/division_covariates.rds"))
age_sex <- read_xlsx(path = "2023-11-28_glasgow-acf.xlsx", sheet = "age_sex_population") %>%
pivot_longer(cols = c(male, female),
names_to = "sex")
#collapse down to smaller age groups to be manageable
age_sex <- age_sex %>%
ungroup() %>%
mutate(age = case_when(age == "0 to 4" ~ "00 to 04",
age == "5 to 9" ~ "05 to 14",
age == "10 to 14" ~ "05 to 14",
age == "15 to 19" ~ "15 to 24",
age == "20 to 24" ~ "15 to 24",
age == "25 to 29" ~ "25 to 34",
age == "30 to 34" ~ "25 to 34",
age == "35 to 39" ~ "35 to 44",
age == "40 to 44" ~ "35 to 44",
age == "45 to 49" ~ "45 to 59",
age == "50 to 54" ~ "45 to 59",
age == "55 to 59" ~ "45 to 59",
TRUE ~ "60 & up")) %>%
group_by(year, age, sex) %>%
mutate(value = sum(value)) %>%
ungroup()
m_age_sex <- lm(value ~ splines::ns(year, knots = 3)*age*sex, data = age_sex)
summary(m_age_sex)
Warning: essentially perfect fit: summary may be unreliable
Call:
lm(formula = value ~ splines::ns(year, knots = 3) * age * sex,
data = age_sex)
Residuals:
Min 1Q Median 3Q Max
-2.107e-10 -7.560e-13 0.000e+00 0.000e+00 2.107e-10
Coefficients: (14 not defined because of singularities)
Estimate Std. Error t value Pr(>|t|)
(Intercept) 5.222e+04 3.820e-10 1.367e+14 <2e-16
splines::ns(year, knots = 3)1 -8.043e+03 7.621e-10 -1.055e+13 <2e-16
splines::ns(year, knots = 3)2 NA NA NA NA
age05 to 14 3.669e+04 4.679e-10 7.843e+13 <2e-16
age15 to 24 -3.893e+03 4.679e-10 -8.320e+12 <2e-16
age25 to 34 -3.996e+04 4.679e-10 -8.540e+13 <2e-16
age35 to 44 -4.230e+04 4.679e-10 -9.040e+13 <2e-16
age45 to 59 5.459e+04 4.411e-10 1.238e+14 <2e-16
age60 & up 7.533e+04 4.126e-10 1.826e+14 <2e-16
sexmale 3.374e+03 5.402e-10 6.244e+12 <2e-16
splines::ns(year, knots = 3)1:age05 to 14 -1.863e+03 9.334e-10 -1.996e+12 <2e-16
splines::ns(year, knots = 3)2:age05 to 14 NA NA NA NA
splines::ns(year, knots = 3)1:age15 to 24 7.533e+04 9.334e-10 8.070e+13 <2e-16
splines::ns(year, knots = 3)2:age15 to 24 NA NA NA NA
splines::ns(year, knots = 3)1:age25 to 34 1.325e+05 9.334e-10 1.420e+14 <2e-16
splines::ns(year, knots = 3)2:age25 to 34 NA NA NA NA
splines::ns(year, knots = 3)1:age35 to 44 1.380e+05 9.334e-10 1.479e+14 <2e-16
splines::ns(year, knots = 3)2:age35 to 44 NA NA NA NA
splines::ns(year, knots = 3)1:age45 to 59 3.474e+03 8.800e-10 3.948e+12 <2e-16
splines::ns(year, knots = 3)2:age45 to 59 NA NA NA NA
splines::ns(year, knots = 3)1:age60 & up -8.453e+04 8.232e-10 -1.027e+14 <2e-16
splines::ns(year, knots = 3)2:age60 & up NA NA NA NA
splines::ns(year, knots = 3)1:sexmale -1.994e+03 1.078e-09 -1.850e+12 <2e-16
splines::ns(year, knots = 3)2:sexmale NA NA NA NA
age05 to 14:sexmale 1.053e+04 6.617e-10 1.592e+13 <2e-16
age15 to 24:sexmale 2.352e+04 6.617e-10 3.555e+13 <2e-16
age25 to 34:sexmale 1.355e+04 6.617e-10 2.047e+13 <2e-16
age35 to 44:sexmale -1.727e+03 6.617e-10 -2.611e+12 <2e-16
age45 to 59:sexmale 2.774e+03 6.238e-10 4.446e+12 <2e-16
age60 & up:sexmale -7.761e+04 5.835e-10 -1.330e+14 <2e-16
splines::ns(year, knots = 3)1:age05 to 14:sexmale -2.049e+04 1.320e-09 -1.552e+13 <2e-16
splines::ns(year, knots = 3)2:age05 to 14:sexmale NA NA NA NA
splines::ns(year, knots = 3)1:age15 to 24:sexmale -6.780e+04 1.320e-09 -5.136e+13 <2e-16
splines::ns(year, knots = 3)2:age15 to 24:sexmale NA NA NA NA
splines::ns(year, knots = 3)1:age25 to 34:sexmale -3.804e+04 1.320e-09 -2.882e+13 <2e-16
splines::ns(year, knots = 3)2:age25 to 34:sexmale NA NA NA NA
splines::ns(year, knots = 3)1:age35 to 44:sexmale -1.171e+04 1.320e-09 -8.874e+12 <2e-16
splines::ns(year, knots = 3)2:age35 to 44:sexmale NA NA NA NA
splines::ns(year, knots = 3)1:age45 to 59:sexmale -3.473e+04 1.244e-09 -2.791e+13 <2e-16
splines::ns(year, knots = 3)2:age45 to 59:sexmale NA NA NA NA
splines::ns(year, knots = 3)1:age60 & up:sexmale 1.056e+05 1.164e-09 9.071e+13 <2e-16
splines::ns(year, knots = 3)2:age60 & up:sexmale NA NA NA NA
(Intercept) ***
splines::ns(year, knots = 3)1 ***
splines::ns(year, knots = 3)2
age05 to 14 ***
age15 to 24 ***
age25 to 34 ***
age35 to 44 ***
age45 to 59 ***
age60 & up ***
sexmale ***
splines::ns(year, knots = 3)1:age05 to 14 ***
splines::ns(year, knots = 3)2:age05 to 14
splines::ns(year, knots = 3)1:age15 to 24 ***
splines::ns(year, knots = 3)2:age15 to 24
splines::ns(year, knots = 3)1:age25 to 34 ***
splines::ns(year, knots = 3)2:age25 to 34
splines::ns(year, knots = 3)1:age35 to 44 ***
splines::ns(year, knots = 3)2:age35 to 44
splines::ns(year, knots = 3)1:age45 to 59 ***
splines::ns(year, knots = 3)2:age45 to 59
splines::ns(year, knots = 3)1:age60 & up ***
splines::ns(year, knots = 3)2:age60 & up
splines::ns(year, knots = 3)1:sexmale ***
splines::ns(year, knots = 3)2:sexmale
age05 to 14:sexmale ***
age15 to 24:sexmale ***
age25 to 34:sexmale ***
age35 to 44:sexmale ***
age45 to 59:sexmale ***
age60 & up:sexmale ***
splines::ns(year, knots = 3)1:age05 to 14:sexmale ***
splines::ns(year, knots = 3)2:age05 to 14:sexmale
splines::ns(year, knots = 3)1:age15 to 24:sexmale ***
splines::ns(year, knots = 3)2:age15 to 24:sexmale
splines::ns(year, knots = 3)1:age25 to 34:sexmale ***
splines::ns(year, knots = 3)2:age25 to 34:sexmale
splines::ns(year, knots = 3)1:age35 to 44:sexmale ***
splines::ns(year, knots = 3)2:age35 to 44:sexmale
splines::ns(year, knots = 3)1:age45 to 59:sexmale ***
splines::ns(year, knots = 3)2:age45 to 59:sexmale
splines::ns(year, knots = 3)1:age60 & up:sexmale ***
splines::ns(year, knots = 3)2:age60 & up:sexmale
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Residual standard error: 5.755e-11 on 44 degrees of freedom
Multiple R-squared: 1, Adjusted R-squared: 1
F-statistic: 1.714e+29 on 27 and 44 DF, p-value: < 2.2e-16
age_levels <- age_sex %>% select(age) %>% distinct() %>% pull()
age_sex_nd <-
crossing(
age=age_levels,
sex=c("male", "female"),
year = 1950:1963
)
pred_pops <- age_sex_nd %>% modelr::add_predictions(m_age_sex)
Warning: prediction from rank-deficient fit; attr(*, "non-estim") has doubtful cases
pred_pops %>%
ggplot(aes(x=year, y=pred, colour=age)) +
geom_line() +
geom_point() +
facet_grid(sex~.) +
scale_y_continuous(labels = comma, limits = c(0, 125000))
#How well do they match up with our overall populations?
pred_pops %>%
group_by(year) %>%
summarise(sum_pred_pop = sum(pred)) %>%
right_join(overall_pops) %>%
select(year, sum_pred_pop, population_without_inst_ship, total_population) %>%
pivot_longer(cols = c(sum_pred_pop, population_without_inst_ship, total_population)) %>%
ggplot(aes(x=year, y=value, colour=name)) +
geom_point() +
scale_y_continuous(labels = comma, limits = c(800000, 1250000))
Joining with `by = join_by(year)`
pred_pops %>%
group_by(year, sex) %>%
summarise(sum = sum(pred)) %>%
group_by(year) %>%
mutate(sex_ratio = first(sum)/last(sum))
`summarise()` has grouped output by 'year'. You can override using the `.groups` argument.
What percentage of adults (15+ participated in the intervention in 1957)?
pred_pops %>%
ungroup() %>%
filter(year==1957) %>%
filter(age != "00 to 04",
age != "05 to 14") %>%
summarise(total_pop = sum(pred)) %>%
mutate(cxr_screened = 622349) %>%
mutate(pct_pop_cxr_screened = percent(cxr_screened/total_pop))
pred_pops %>%
ungroup() %>%
filter(year==1957) %>%
filter(age != "00 to 04",
age != "05 to 14") %>%
summarise(total_pop = sum(pred), .by=sex) %>%
mutate(cxr_screened = c(340474, 281875)) %>%
mutate(pct_pop_cxr_screened = percent(cxr_screened/total_pop))
NA
NA
Population pyramids
label_abs <- function(x) {
comma(abs(x))
}
pred_pops %>%
ungroup() %>%
group_by(year) %>%
mutate(year_pop = sum(pred),
age_sex_pct = percent(pred/year_pop, accuracy=0.1)) %>%
mutate(sex = case_when(sex=="male" ~ "Male",
sex=="female" ~ "Female")) %>%
ggplot(
aes(x = age, fill = sex,
y = ifelse(test = sex == "Female",yes = -pred, no = pred))) +
geom_bar(stat = "identity") +
geom_text(aes(label = age_sex_pct),
position= position_stack(vjust=0.5), colour="white", size=2.5) +
facet_wrap(year~., ncol=7) +
coord_flip() +
scale_y_continuous(labels = label_abs) +
scale_fill_manual(values = c("mediumseagreen", "purple"), name="") +
theme_ggdist() +
theme(axis.text.x = element_text(angle=90, hjust = 1, vjust=0.5),
legend.position = "bottom",
panel.border = element_rect(colour = "grey78", fill=NA)) +
labs(x="", y="")
ggsave(here("figures/s3.png"), width=10)
Saving 10 x 4.51 in image
Not perfect, but resonably good. But ahhhhh… the age groups don’t align with the case notification age groups! Come back to think about this later.
Import the tuberculosis cases dataset
Overall, by year.
cases_by_year <- read_xlsx("2023-11-28_glasgow-acf.xlsx", sheet = "by_year")
cases_by_year%>%
mutate(across(where(is.numeric) & !(year), ~comma(.))) %>%
datatable()
#shift year to midpoint
cases_by_year <- cases_by_year %>%
mutate(year2 = year+0.5)
Plot the overall number of case notified per year, by pulmonary and extra pulmonary classification.
cases_by_year %>%
select(-total_notifications, -year) %>%
pivot_longer(cols = c(pulmonary_notifications, `non-pulmonary_notifications`)) %>%
mutate(name = case_when(name == "pulmonary_notifications" ~ "Pulmonary TB",
name == "non-pulmonary_notifications" ~ "Extra-pulmonary TB")) %>%
ggplot() +
geom_area(aes(y=value, x=year2, group = name, fill=name), alpha=0.5) +
geom_vline(aes(xintercept=acf_start), linetype=3) +
geom_vline(aes(xintercept=acf_end), linetype=3) +
scale_y_continuous(labels=comma) +
scale_x_continuous(labels = year_labels,
breaks = year_labels) +
scale_fill_brewer(palette = "Set1", name="") +
labs(
title = "Glasgow Corporation: Tuberculosis notifications",
subtitle = "1950 to 1963, by TB disease classification",
x = "Year",
y = "Number of cases",
caption = "Mass miniature X-ray campaign period between dashed lines (11th March-12th April 1957)"
) +
theme_ggdist() +
theme(legend.position = "bottom")
NA
NA
Read in the datasets and merge together.
#list all the sheets
all_sheets <- excel_sheets("2023-11-28_glasgow-acf.xlsx")
#get the ward sheets
ward_sheets <- enframe(all_sheets) %>%
filter(grepl("by_ward", value)) %>%
pull(value)
cases_by_ward_sex_year <- map_df(ward_sheets, ~read_xlsx(path = "2023-11-28_glasgow-acf.xlsx",
sheet = .))
cases_by_ward_sex_year %>%
mutate(across(where(is.numeric) & !(year), ~comma(.))) %>%
datatable()
NA
Aggregate together to get cases by division
cases_by_division <- cases_by_ward_sex_year %>%
left_join(ward_lookup) %>%
group_by(division, year, tb_type) %>%
summarise(cases = sum(cases, na.rm = TRUE))
Joining with `by = join_by(ward)``summarise()` has grouped output by 'division', 'year'. You can override using the `.groups` argument.
#shift year to midpoint
cases_by_division <- cases_by_division %>%
mutate(year2 = year+0.5) %>%
ungroup()
cases_by_division %>%
select(-year2) %>%
select(year, everything()) %>%
mutate(across(where(is.numeric) & !(year), ~comma(.))) %>%
datatable()
cases_by_division %>%
mutate(tb_type = case_when(tb_type == "Pulmonary" ~ "Pulmonary TB",
tb_type == "Non-Pulmonary" ~ "Extra-pulmonary TB")) %>%
ggplot() +
geom_area(aes(y=cases, x=year2, group = tb_type, fill=tb_type), alpha=0.5) +
geom_vline(aes(xintercept=acf_start), linetype=3) +
geom_vline(aes(xintercept=acf_end), linetype=3) +
scale_y_continuous(labels=comma) +
scale_x_continuous(labels = year_labels,
breaks = year_labels,
guide = guide_axis(angle = 90)) +
facet_wrap(division~., scales = "free_y") +
scale_fill_brewer(palette = "Set1", name="") +
labs(
title = "Glasgow Corporation: Tuberculosis notifications by Division",
subtitle = "1950 to 1963, by TB disease classification",
x = "Year",
y = "Number of cases",
caption = "Mass miniature X-ray campaign period between dashed lines (11th March-12th April 1957)\nNote: extra-pulmonary TB cases by Division/Ward not reported in 1962-1963"
) +
theme_ggdist() +
theme(legend.position = "bottom")
cases_by_ward <- cases_by_ward_sex_year %>%
group_by(ward, year, tb_type) %>%
summarise(cases = sum(cases, na.rm = TRUE)) %>%
ungroup()
`summarise()` has grouped output by 'ward', 'year'. You can override using the `.groups` argument.
cases_by_ward %>%
mutate(across(where(is.numeric) & !(year), ~comma(.))) %>%
select(year, everything()) %>%
datatable()
#shift year to midpoint
cases_by_ward <- cases_by_ward %>%
mutate(year2 = year+0.5)
cases_by_ward %>%
mutate(tb_type = case_when(tb_type == "Pulmonary" ~ "Pulmonary TB",
tb_type == "Non-Pulmonary" ~ "Extra-pulmonary TB")) %>%
ggplot() +
geom_area(aes(y=cases, x=year2, group = tb_type, fill=tb_type), alpha=0.8) +
geom_vline(aes(xintercept=acf_start), linetype=3) +
geom_vline(aes(xintercept=acf_end), linetype=3) +
scale_y_continuous(labels=comma) +
scale_x_continuous(labels = year_labels,
breaks = year_labels,
guide = guide_axis(angle = 90)) +
facet_wrap(ward~., scales = "free_y") +
scale_fill_brewer(palette = "Set1", name="") +
labs(
title = "Glasgow Corporation: Tuberculosis notifications by Ward",
subtitle = "1950 to 1963, by TB disease classification",
x = "Year",
y = "Number of cases",
caption = "Mass miniature X-ray campaign period between dashed lines (11th March-12th April 1957)\nNote: extra-pulmonary TB cases by Division/Ward not reported in 1962-1963"
) +
theme(legend.position = "bottom")
NA
NA
As we don’t have denominators, we will just model the change in counts.
#list all the sheets
all_sheets <- excel_sheets("2023-11-28_glasgow-acf.xlsx")
#get the ward sheets
age_sex_sheets <- enframe(all_sheets) %>%
filter(grepl("by_age_sex", value)) %>%
pull(value)
cases_by_age_sex <- map_df(age_sex_sheets, ~read_xlsx(path = "2023-11-28_glasgow-acf.xlsx",
sheet = .))
cases_by_age_sex %>%
mutate(across(where(is.numeric) & !(year), ~comma(.))) %>%
datatable()
NA
NA
Now calculate case notification rates per 100,000 population
Merge the notification and population denominator datasets together.
Here we need to include the whole population (with shipping and institutions) as they are included in the notifications.
overall_inc <- overall_pops %>%
left_join(cases_by_year)
Joining with `by = join_by(year, year2)`
overall_inc <- overall_inc %>%
mutate(inc_pulm_100k = pulmonary_notifications/total_population*100000,
inc_ep_100k = `non-pulmonary_notifications`/total_population*100000,
inc_100k = total_notifications/total_population*100000)
overall_inc %>%
select(year, inc_100k, inc_pulm_100k, inc_ep_100k) %>%
mutate_at(.vars = vars(inc_100k, inc_pulm_100k, inc_ep_100k),
.funs = funs(round)) %>%
datatable()
Warning: `funs()` was deprecated in dplyr 0.8.0.
Please use a list of either functions or lambdas:
# Simple named list:
list(mean = mean, median = median)
# Auto named with `tibble::lst()`:
tibble::lst(mean, median)
# Using lambdas
list(~ mean(., trim = .2), ~ median(., na.rm = TRUE))
overall_inc %>%
select(year2, inc_pulm_100k, inc_ep_100k) %>%
pivot_longer(cols = c(inc_pulm_100k, `inc_ep_100k`)) %>%
mutate(name = case_when(name == "inc_pulm_100k" ~ "Pulmonary TB",
name == "inc_ep_100k" ~ "Extra-pulmonary TB")) %>%
ggplot() +
geom_area(aes(y=value, x=year2, group = name, fill=name), alpha=0.5) +
geom_vline(aes(xintercept=acf_start), linetype=3) +
geom_vline(aes(xintercept=acf_end), linetype=3) +
scale_y_continuous(labels=comma) +
scale_x_continuous(labels = year_labels,
breaks = year_labels) +
scale_fill_brewer(palette = "Set1", name="") +
labs(
title = "Glasgow Corporation: Tuberculosis case notification rate",
subtitle = "1950 to 1963, by TB disease classification",
x = "Year",
y = "Case notification rate (per 100,000)",
caption = "Mass miniature X-ray campaign period between dashed lines (11th March-12th April 1957)"
) +
theme_ggdist() +
theme(legend.position = "bottom")
NA
NA
NA
division_inc <- division_pops %>%
left_join(cases_by_division)
Joining with `by = join_by(division, year)`
division_inc <- division_inc %>%
mutate(inc_100k = cases/total_population*100000)
division_inc %>%
select(year, division, tb_type, inc_100k) %>%
mutate_at(.vars = vars(inc_100k),
.funs = funs(round)) %>%
datatable()
Warning: `funs()` was deprecated in dplyr 0.8.0.
Please use a list of either functions or lambdas:
# Simple named list:
list(mean = mean, median = median)
# Auto named with `tibble::lst()`:
tibble::lst(mean, median)
# Using lambdas
list(~ mean(., trim = .2), ~ median(., na.rm = TRUE))
division_inc %>%
mutate(tb_type = case_when(tb_type == "Pulmonary" ~ "Pulmonary TB",
tb_type == "Non-Pulmonary" ~ "Extra-pulmonary TB")) %>%
ggplot() +
geom_area(aes(y=inc_100k, x=year2, group = tb_type, fill=tb_type), alpha=0.5) +
geom_vline(aes(xintercept=acf_start), linetype=3) +
geom_vline(aes(xintercept=acf_end), linetype=3) +
scale_y_continuous(labels=comma) +
scale_x_continuous(labels = year_labels,
breaks = year_labels,
guide = guide_axis(angle = 90)) +
facet_wrap(division~.) +
scale_fill_brewer(palette = "Set1", name="") +
labs(
title = "Glasgow Corporation: Tuberculosis case notification rate, by Division",
subtitle = "1950 to 1963, by TB disease classification",
x = "Year",
y = "Case notification rate (per 100,000)",
caption = "Mass miniature X-ray campaign period between dashed lines (11th March-12th April 1957)\nNote: extra-pulmonary TB cases by Division/Ward not reported in 1962-1963"
) +
theme_ggdist() +
theme(legend.position = "bottom")
NA
NA
NA
Here we will filter out the institutions and harbour from the denominators, as we don’t have reliable population denominators for them.
ward_inc <- ward_pops %>%
left_join(cases_by_ward)
Joining with `by = join_by(ward, year, year2)`
ward_inc <- ward_inc %>%
mutate(inc_100k = cases/population_without_inst_ship*100000)
ward_inc %>%
select(year, ward, tb_type, inc_100k) %>%
mutate_at(.vars = vars(inc_100k),
.funs = funs(round)) %>%
datatable()
Warning: `funs()` was deprecated in dplyr 0.8.0.
Please use a list of either functions or lambdas:
# Simple named list:
list(mean = mean, median = median)
# Auto named with `tibble::lst()`:
tibble::lst(mean, median)
# Using lambdas
list(~ mean(., trim = .2), ~ median(., na.rm = TRUE))
ward_inc %>%
mutate(tb_type = case_when(tb_type == "Pulmonary" ~ "Pulmonary TB",
tb_type == "Non-Pulmonary" ~ "Extra-pulmonary TB")) %>%
ggplot() +
geom_area(aes(y=inc_100k, x=year2, group = tb_type, fill=tb_type), alpha=0.5) +
geom_vline(aes(xintercept=acf_start), linetype=3) +
geom_vline(aes(xintercept=acf_end), linetype=3) +
scale_y_continuous(labels=comma) +
scale_x_continuous(labels = year_labels,
breaks = year_labels,
guide = guide_axis(angle = 90)) +
facet_wrap(ward~.) +
scale_fill_brewer(palette = "Set1", name="") +
labs(
title = "Glasgow Corporation: Tuberculosis case notification rate, by Ward",
subtitle = "1950 to 1963, by TB disease classification",
x = "Year",
y = "Incidence (per 100,000)",
caption = "Mass miniature X-ray campaign period between dashed lines (11th March-12th April 1957)\nNote: extra-pulmonary TB cases by Division/Ward not reported in 1962-1963"
) +
theme(legend.position = "bottom")
NA
NA
NA
NA
On a map
st_as_sf(left_join(ward_inc, glasgow_wards_1951)) %>%
filter(tb_type=="Pulmonary") %>%
ggplot() +
geom_sf(aes(fill=inc_100k)) +
facet_wrap(year~., ncol = 7) +
scale_fill_viridis_c(name="Case notification rate (per 100,000)",
option = "A") +
theme_ggdist() +
theme(legend.position = "top",
legend.key.width = unit(2, "cm"),
panel.border = element_rect(colour = "grey78", fill=NA)) +
guides(fill=guide_colorbar(title.position = "top"))
Joining with `by = join_by(division, ward, ward_number)`
Import the TB mortality data.
First, overall deaths. Note that in the original reports, we have a pulmonary TB death rate per million for all years, and numbers of pulmonary TB deaths for each year apart from 1950.
#get the overall mortality sheets
deaths_sheets <- enframe(all_sheets) %>%
filter(grepl("deaths", value)) %>%
pull(value)
overall_deaths <- map_df(deaths_sheets, ~read_xlsx(path = "2023-11-28_glasgow-acf.xlsx",
sheet = .))
overall_deaths %>%
mutate(across(where(is.numeric) & !(year), ~comma(.))) %>%
datatable()
NA
NA
NA
Plot the raw numbers of pulmonary deaths
overall_deaths %>%
ggplot(aes(x=year, y=pulmonary_deaths)) +
geom_line(colour = "#DE0D92") +
geom_point(colour = "#DE0D92") +
geom_vline(aes(xintercept=acf_start), linetype=3) +
geom_vline(aes(xintercept=acf_end), linetype=3) +
labs(y="Pulmonary TB deaths per year",
x = "Year",
title = "Numbers of pulmonary TB deaths",
subtitle = "Glasgow, 1950-1963",
caption = "Mass miniature X-ray campaign period between dashed lines (11th March-12th April 1957)\nNote: no data for 1950") +
theme_ggdist() +
theme(panel.border = element_rect(colour = "grey78", fill=NA))
NA
NA
Now the incidence of pulmonary TB death
overall_deaths %>%
ggplot(aes(x=year, y=pulmonary_death_rate_per_100k)) +
geom_line(colour = "#4D6CFA") +
geom_point(colour = "#4D6CFA") +
geom_vline(aes(xintercept=acf_start), linetype=3) +
geom_vline(aes(xintercept=acf_end), linetype=3) +
scale_y_continuous(labels=comma) +
scale_x_continuous(labels = year_labels,
breaks = year_labels) +
labs(y="Annual incidence of death (per 100,000)",
x = "Year",
caption = "Mass miniature X-ray campaign period between dashed lines (11th March-12th April 1957)") +
theme_ggdist() +
theme(panel.border = element_rect(colour = "grey78", fill=NA))
ggsave(here("figures/s7.png"), width=10)
Saving 10 x 4.51 in image
Make Table 1 here, and save for publication.
overall_pops %>%
select(year, total_population) %>%
left_join(overall_inc %>%
select(year,
pulmonary_notifications, inc_pulm_100k,
`non-pulmonary_notifications`, inc_ep_100k,
total_notifications, inc_100k)) %>%
left_join(overall_deaths %>%
select(year,
pulmonary_deaths, pulmonary_death_rate_per_100k)) %>%
mutate(across(where(is.numeric) & !(year), ~round(., digits=1))) %>%
mutate(across(where(is.numeric) & !(year), ~comma(.)))
Joining with `by = join_by(year)`Joining with `by = join_by(year)`
Prepare the datasets for modelling
mdata <- ward_inc %>%
filter(tb_type=="Pulmonary") %>%
mutate(acf_period = case_when(year %in% c(1950:1956) ~ "a. pre-acf",
year %in% c(1957) ~ "b. acf",
year %in% c(1958:1963) ~ "c. post-acf")) %>%
group_by(ward) %>%
mutate(y_num = row_number()) %>%
ungroup()
mdata_extrapulmonary <- ward_inc %>%
filter(tb_type=="Non-Pulmonary") %>%
mutate(acf_period = case_when(year %in% c(1950:1956) ~ "a. pre-acf",
year %in% c(1957) ~ "b. acf",
year %in% c(1958:1963) ~ "c. post-acf")) %>%
group_by(ward) %>%
mutate(y_num = row_number()) %>%
ungroup() %>%
filter(year<=1961) #no data for 1962 and 1963
#scaffold for overall predictions
overall_scaffold <- mdata %>%
select(year, year2, y_num, acf_period, population_without_inst_ship, ward, cases) %>%
group_by(year, year2, y_num, acf_period) %>%
summarise(population_without_inst_ship = sum(population_without_inst_ship),
cases = sum(cases)) %>%
ungroup() %>%
mutate(inc_100k = cases/population_without_inst_ship*100000) %>%
left_join(mdata_extrapulmonary %>% group_by(year) %>%
summarise(cases_extrapulmonary = sum(cases))) %>%
mutate(inc_100k_extrapulmonary = cases_extrapulmonary/population_without_inst_ship*100000)
`summarise()` has grouped output by 'year', 'year2', 'y_num'. You can override using the `.groups` argument.Joining with `by = join_by(year)`
This models the case notification rate over time, with a step change for the intervention, and slope change after the intervention.
Work on the priors a bit. We will build up from less complex to more complex.
at the intercept, we expect somewhere around 2500. We will set the standard deviation to both 0.5 and 1 to check what it looks like
#
# c(prior(lognormal(7.600902, 0.5)), #log(2500) = 7.600902
# prior(lognormal(7.600902, 1))) %>%
# parse_dist() %>%
#
# ggplot(aes(y = prior, dist = .dist, args = .args)) +
# stat_halfeye(.width = c(.5, .95)) +
# scale_y_discrete(NULL, labels = str_c("lognormal(log(2000), ", c(0.5, 1), ")"),
# expand = expansion(add = 0.1)) +
# xlab(expression(exp(italic(p)(beta[0])))) +
# coord_cartesian(xlim = c(0,15000))
#
#
# prior(gamma(1, 0.01)) %>%
# parse_dist() %>%
# ggplot(aes(y=prior, dist = .dist, args = .args)) +
# stat_halfeye(.width = c(0.5, 0.95))
#
# #now fit to a model, and plot some prior realisations
#
# m_prior1 <- brm(
# cases ~ 0 + Intercept,
# family = negbinomial(),
# data = overall_scaffold,
# sample_prior = "only",
# prior = prior(normal(log(2000), 0.5), class = b, coef = Intercept) +
# prior(gamma(1, 0.01), class = shape)
# )
#
# add_epred_draws(object=m_prior1,
# newdata = tibble(intercept=1)) %>%
# ggplot(aes(x=intercept, y=.epred)) +
# stat_halfeye() +
# scale_y_log10(labels = comma)
Now try to add in a term for the effect of y_num. We anticpate that the number of cases will decline by about 1-5% per year. However, as we are pretty uncertain about this, we will just encode a weakly regularising prior to restrict the year size to sensible ranges.
#
#
# m_prior2 <- brm(
# cases ~ 0 + Intercept + y_num,
# family = negbinomial(),
# data = overall_scaffold,
# sample_prior = "only",
# prior = prior(normal(log(2000), 0.5), class = b, coef = Intercept) +
# prior(gamma(1, 0.01), class = shape) +
# prior(normal(0, 0.01), class = b, coef = y_num)
# )
#
# add_epred_draws(object=m_prior2,
# newdata = overall_scaffold) %>%
# ggplot(aes(x=year, y=.epred)) +
# stat_halfeye() +
# scale_y_log10(label=comma)
Now we want to add in a prior for the effect of the acf_intervention. We anticipate the peak to be anywhere between no effect, and a tripling
#
# m_prior3 <- brm(
# cases ~ 0 + Intercept + y_num + acf_period,
# family = negbinomial(),
# data = overall_scaffold,
# sample_prior = "only",
# prior = prior(normal(log(2000), 0.5), class = b, coef = Intercept) +
# prior(gamma(1, 0.01), class = shape) +
# prior(normal(0, 0.01), class = b, coef = y_num) +
# prior(normal(0, 0.001), class = b)
# )
#
#
# add_epred_draws(object=m_prior3,
# newdata = overall_scaffold) %>%
# ggplot(aes(x=year, y=.epred)) +
# stat_halfeye() +
# scale_y_log10(labels = comma)
#
Now we look and see what it looks like with the interactions
#
# m_prior4 <- brm(
# cases ~ 0 + Intercept + y_num + acf_period + y_num:acf_period,
# family = negbinomial(),
# data = overall_scaffold,
# sample_prior = "only",
# prior = prior(normal(log(2500), 1), class = b, coef = Intercept) +
# prior(gamma(1, 0.01), class = shape) +
# prior(normal(0, 0.01), class = b)
# )
#
# add_epred_draws(object=m_prior4,
# newdata = overall_scaffold) %>%
# ggplot(aes(x=year, y=.epred)) +
# stat_halfeye() +
# scale_y_log10(label=comma)
#
#
Now try adding in the random intercepts
# c(prior(lognormal(3.912023, 0.5)), #log(50) = 3.912023
# prior(lognormal(3.912023, 1))) %>%
# parse_dist() %>%
#
# ggplot(aes(y = prior, dist = .dist, args = .args)) +
# stat_halfeye(.width = c(.5, .95)) +
# scale_y_discrete(NULL, labels = str_c("lognormal(log(50), ", c(0.5, 1), ")"),
# expand = expansion(add = 0.1)) +
# xlab(expression(exp(italic(p)(beta[0])))) +
# coord_cartesian(xlim = c(0,400))
#
#
# m_prior5 <- brm(
# cases ~ y_num + acf_period + y_num:acf_period + ( 1 | ward),
# family = negbinomial(),
# data = mdata,
# sample_prior = "only",
# prior = prior(normal(log(50), 1), class = Intercept) +
# prior(gamma(1, 0.01), class = shape) +
# prior(normal(0, 0.01), class = b) +
# prior(exponential(1), class=sd)
# )
#
#
# add_epred_draws(object=m_prior5,
# newdata = mdata,
# re_formula = NA) %>%
# ggplot(aes(x=year, y=.epred)) +
# stat_halfeye() +
# scale_y_log10(label=comma)
#
# add_epred_draws(object=m_prior5,
# newdata = mdata,
# re_formula = NA) %>%
# ggplot(aes(x=year, y=.epred)) +
# stat_halfeye() +
# scale_y_log10(label=comma) +
# facet_wrap(ward~.)
And add in the random slopes
#
# m_prior6 <- brm(
# cases ~ 1 + y_num + acf_period + y_num:acf_period + (1 + y_num*acf_period | ward),
# family = negbinomial(),
# data = mdata,
# sample_prior = "only",
# prior = prior(gamma(1, 0.01), class = shape) +
# prior(normal(0, 0.1), class = b) +
# prior(exponential(1), class=sd) +
# prior(lkj(2), class=cor)
# )
#
#
#
# m_prior6 <- brm(
# cases ~ 0 + Intercept + y_num + acf_period + y_num:acf_period + ( y_num*acf_period | ward),
# family = negbinomial(),
# data = mdata,
# sample_prior = "only",
# prior = prior(normal(log(50), 1), class = b, coef = Intercept) +
# prior(gamma(1, 0.01), class = shape) +
# prior(normal(0, 0.01), class = b) +
# prior(exponential(100), class=sd) +
# prior(lkj(2), class=cor)
# )
# add_epred_draws(object=m_prior6,
# newdata = mdata,
# re_formula = NA) %>%
# ggplot(aes(x=year, y=.epred)) +
# stat_halfeye() +
# scale_y_log10(label=comma)
#
# add_epred_draws(object=m_prior6,
# newdata = mdata,
# re_formula = ~( 1 + y_num + acf_period | ward)) %>%
# ggplot(aes(x=year, y=.epred)) +
# stat_halfeye() +
# scale_y_log10(label=comma) +
# facet_wrap(ward~.)
#
# plot_counterfactual(model_data = overall_scaffold, model=m_prior6, outcome = inc_100k,
# population_denominator = population_without_inst_ship, re_formula = NA)
#
# plot_counterfactual(model_data = mdata, model=m_prior6, outcome = inc_100k,
# population_denominator = population_without_inst_ship, grouping_var = ward, ward,
# re_formula = ~( 1 + y_num + acf_period | ward))
Issue here is the non-centred parameterisation of the intercept prior… Feel like this is a more interpretable way to set priors… but will revert to centred parameterisation for the meantime.
# m_centered_prior <- brm(
# cases ~ 1 + y_num*acf_period + (1 + y_num*acf_period | ward) + offset(log(population_without_inst_ship)),
# data = mdata,
# family = negbinomial(),
# seed = 1234,
# chains = 4, cores = 4,
# prior = prior(normal(0,1000), class = Intercept) +
# prior(gamma(0.01, 0.01), class = shape) +
# prior(normal(0, 1), class = b) +
# prior(exponential(1), class=sd) +
# prior(lkj(2), class=cor),
# sample_prior = "only")
#
# plot(m_centered_prior)
#
# plot_counterfactual(model_data = overall_scaffold, model=m_centered_prior, outcome = inc_100k,
# population_denominator = population_without_inst_ship, re_formula = NA)
#
# plot_counterfactual(model_data = mdata, model=m_centered_prior, outcome = inc_100k,
# population_denominator = population_without_inst_ship, grouping_var = ward, ward,
# re_formula = ~( 1 + y_num*acf_period | ward))
Look at the mean and variance of counts (counts of pulmonary notifications are what we are predicting)
#Mean of counts per year
mean(mdata$cases)
[1] 48.32819
#variance of counts per year
var(mdata$cases)
[1] 915.5749
Quite a bit of over-dispersion here, so negative binomial distribution might be a better choice of distributional family than Poisson.
Fit the model with the data
m_pulmonary <- brm(
cases ~ 1 + y_num*acf_period + (1 + y_num*acf_period | ward) + offset(log(population_without_inst_ship)),
data = mdata,
family = negbinomial(),
seed = 1234,
chains = 4, cores = 4,
prior = prior(normal(0,1), class = Intercept) +
prior(gamma(0.01, 0.01), class = shape) +
prior(normal(0, 1), class = b) +
prior(exponential(1), class=sd) +
prior(lkj(4), class=cor),
control = list(adapt_delta = 0.9))
Compiling Stan program...
Start sampling
starting worker pid=10072 on localhost:11273 at 19:31:46.486
starting worker pid=10085 on localhost:11273 at 19:31:46.579
starting worker pid=10098 on localhost:11273 at 19:31:46.669
starting worker pid=10111 on localhost:11273 at 19:31:46.760
SAMPLING FOR MODEL 'anon_model' NOW (CHAIN 1).
Chain 1:
Chain 1: Gradient evaluation took 0.000147 seconds
Chain 1: 1000 transitions using 10 leapfrog steps per transition would take 1.47 seconds.
Chain 1: Adjust your expectations accordingly!
Chain 1:
Chain 1:
Chain 1: Iteration: 1 / 2000 [ 0%] (Warmup)
SAMPLING FOR MODEL 'anon_model' NOW (CHAIN 2).
Chain 2:
Chain 2: Gradient evaluation took 0.000173 seconds
Chain 2: 1000 transitions using 10 leapfrog steps per transition would take 1.73 seconds.
Chain 2: Adjust your expectations accordingly!
Chain 2:
Chain 2:
Chain 2: Iteration: 1 / 2000 [ 0%] (Warmup)
SAMPLING FOR MODEL 'anon_model' NOW (CHAIN 3).
Chain 3:
Chain 3: Gradient evaluation took 0.000149 seconds
Chain 3: 1000 transitions using 10 leapfrog steps per transition would take 1.49 seconds.
Chain 3: Adjust your expectations accordingly!
Chain 3:
Chain 3:
Chain 3: Iteration: 1 / 2000 [ 0%] (Warmup)
SAMPLING FOR MODEL 'anon_model' NOW (CHAIN 4).
Chain 4:
Chain 4: Gradient evaluation took 0.000169 seconds
Chain 4: 1000 transitions using 10 leapfrog steps per transition would take 1.69 seconds.
Chain 4: Adjust your expectations accordingly!
Chain 4:
Chain 4:
Chain 4: Iteration: 1 / 2000 [ 0%] (Warmup)
Chain 1: Iteration: 200 / 2000 [ 10%] (Warmup)
Chain 3: Iteration: 200 / 2000 [ 10%] (Warmup)
Chain 4: Iteration: 200 / 2000 [ 10%] (Warmup)
Chain 2: Iteration: 200 / 2000 [ 10%] (Warmup)
Chain 4: Iteration: 400 / 2000 [ 20%] (Warmup)
Chain 1: Iteration: 400 / 2000 [ 20%] (Warmup)
Chain 2: Iteration: 400 / 2000 [ 20%] (Warmup)
Chain 3: Iteration: 400 / 2000 [ 20%] (Warmup)
Chain 4: Iteration: 600 / 2000 [ 30%] (Warmup)
Chain 3: Iteration: 600 / 2000 [ 30%] (Warmup)
Chain 2: Iteration: 600 / 2000 [ 30%] (Warmup)
Chain 1: Iteration: 600 / 2000 [ 30%] (Warmup)
Chain 4: Iteration: 800 / 2000 [ 40%] (Warmup)
Chain 3: Iteration: 800 / 2000 [ 40%] (Warmup)
Chain 2: Iteration: 800 / 2000 [ 40%] (Warmup)
Chain 1: Iteration: 800 / 2000 [ 40%] (Warmup)
Chain 4: Iteration: 1000 / 2000 [ 50%] (Warmup)
Chain 4: Iteration: 1001 / 2000 [ 50%] (Sampling)
Chain 3: Iteration: 1000 / 2000 [ 50%] (Warmup)
Chain 3: Iteration: 1001 / 2000 [ 50%] (Sampling)
Chain 2: Iteration: 1000 / 2000 [ 50%] (Warmup)
Chain 2: Iteration: 1001 / 2000 [ 50%] (Sampling)
Chain 1: Iteration: 1000 / 2000 [ 50%] (Warmup)
Chain 1: Iteration: 1001 / 2000 [ 50%] (Sampling)
Chain 4: Iteration: 1200 / 2000 [ 60%] (Sampling)
Chain 3: Iteration: 1200 / 2000 [ 60%] (Sampling)
Chain 2: Iteration: 1200 / 2000 [ 60%] (Sampling)
Chain 1: Iteration: 1200 / 2000 [ 60%] (Sampling)
Chain 4: Iteration: 1400 / 2000 [ 70%] (Sampling)
Chain 3: Iteration: 1400 / 2000 [ 70%] (Sampling)
Chain 2: Iteration: 1400 / 2000 [ 70%] (Sampling)
Chain 1: Iteration: 1400 / 2000 [ 70%] (Sampling)
Chain 4: Iteration: 1600 / 2000 [ 80%] (Sampling)
Chain 3: Iteration: 1600 / 2000 [ 80%] (Sampling)
Chain 2: Iteration: 1600 / 2000 [ 80%] (Sampling)
Chain 1: Iteration: 1600 / 2000 [ 80%] (Sampling)
Chain 4: Iteration: 1800 / 2000 [ 90%] (Sampling)
Chain 3: Iteration: 1800 / 2000 [ 90%] (Sampling)
Chain 2: Iteration: 1800 / 2000 [ 90%] (Sampling)
Chain 1: Iteration: 1800 / 2000 [ 90%] (Sampling)
Chain 4: Iteration: 2000 / 2000 [100%] (Sampling)
Chain 4:
Chain 4: Elapsed Time: 22.953 seconds (Warm-up)
Chain 4: 18.619 seconds (Sampling)
Chain 4: 41.572 seconds (Total)
Chain 4:
Chain 3: Iteration: 2000 / 2000 [100%] (Sampling)
Chain 3:
Chain 3: Elapsed Time: 24.347 seconds (Warm-up)
Chain 3: 18.372 seconds (Sampling)
Chain 3: 42.719 seconds (Total)
Chain 3:
Chain 2: Iteration: 2000 / 2000 [100%] (Sampling)
Chain 2:
Chain 2: Elapsed Time: 24.652 seconds (Warm-up)
Chain 2: 18.47 seconds (Sampling)
Chain 2: 43.122 seconds (Total)
Chain 2:
Chain 1: Iteration: 2000 / 2000 [100%] (Sampling)
Chain 1:
Chain 1: Elapsed Time: 25.503 seconds (Warm-up)
Chain 1: 18.491 seconds (Sampling)
Chain 1: 43.994 seconds (Total)
Chain 1:
#check model diagnostics
summary(m_pulmonary)
Family: negbinomial
Links: mu = log; shape = identity
Formula: cases ~ 1 + y_num * acf_period + (1 + y_num * acf_period | ward) + offset(log(population_without_inst_ship))
Data: mdata (Number of observations: 518)
Draws: 4 chains, each with iter = 2000; warmup = 1000; thin = 1;
total post-warmup draws = 4000
Group-Level Effects:
~ward (Number of levels: 37)
Estimate Est.Error l-95% CI u-95% CI
sd(Intercept) 0.25 0.03 0.19 0.32
sd(y_num) 0.02 0.01 0.01 0.03
sd(acf_periodb.acf) 0.06 0.04 0.00 0.16
sd(acf_periodc.postMacf) 0.11 0.06 0.01 0.24
sd(y_num:acf_periodb.acf) 0.01 0.01 0.00 0.02
sd(y_num:acf_periodc.postMacf) 0.01 0.01 0.00 0.02
cor(Intercept,y_num) -0.46 0.20 -0.76 -0.00
cor(Intercept,acf_periodb.acf) -0.23 0.28 -0.71 0.36
cor(y_num,acf_periodb.acf) -0.04 0.27 -0.54 0.48
cor(Intercept,acf_periodc.postMacf) -0.14 0.24 -0.59 0.35
cor(y_num,acf_periodc.postMacf) 0.10 0.25 -0.40 0.57
cor(acf_periodb.acf,acf_periodc.postMacf) 0.06 0.28 -0.49 0.57
cor(Intercept,y_num:acf_periodb.acf) -0.25 0.29 -0.72 0.37
cor(y_num,y_num:acf_periodb.acf) -0.04 0.27 -0.55 0.49
cor(acf_periodb.acf,y_num:acf_periodb.acf) -0.06 0.28 -0.59 0.49
cor(acf_periodc.postMacf,y_num:acf_periodb.acf) 0.07 0.27 -0.48 0.57
cor(Intercept,y_num:acf_periodc.postMacf) -0.02 0.26 -0.50 0.49
cor(y_num,y_num:acf_periodc.postMacf) -0.03 0.27 -0.54 0.51
cor(acf_periodb.acf,y_num:acf_periodc.postMacf) 0.05 0.28 -0.50 0.57
cor(acf_periodc.postMacf,y_num:acf_periodc.postMacf) -0.08 0.29 -0.63 0.47
cor(y_num:acf_periodb.acf,y_num:acf_periodc.postMacf) 0.05 0.27 -0.49 0.55
Rhat Bulk_ESS Tail_ESS
sd(Intercept) 1.00 1465 2204
sd(y_num) 1.00 1072 1047
sd(acf_periodb.acf) 1.00 2045 2257
sd(acf_periodc.postMacf) 1.00 913 1346
sd(y_num:acf_periodb.acf) 1.01 1233 1598
sd(y_num:acf_periodc.postMacf) 1.01 709 1430
cor(Intercept,y_num) 1.00 2716 2233
cor(Intercept,acf_periodb.acf) 1.00 3541 2755
cor(y_num,acf_periodb.acf) 1.00 6229 3130
cor(Intercept,acf_periodc.postMacf) 1.00 5511 2516
cor(y_num,acf_periodc.postMacf) 1.00 2979 3082
cor(acf_periodb.acf,acf_periodc.postMacf) 1.00 2417 2633
cor(Intercept,y_num:acf_periodb.acf) 1.00 3650 3281
cor(y_num,y_num:acf_periodb.acf) 1.00 6105 3309
cor(acf_periodb.acf,y_num:acf_periodb.acf) 1.00 5303 3005
cor(acf_periodc.postMacf,y_num:acf_periodb.acf) 1.00 4234 3467
cor(Intercept,y_num:acf_periodc.postMacf) 1.00 5928 2788
cor(y_num,y_num:acf_periodc.postMacf) 1.00 4067 3118
cor(acf_periodb.acf,y_num:acf_periodc.postMacf) 1.00 2699 3112
cor(acf_periodc.postMacf,y_num:acf_periodc.postMacf) 1.00 3352 3170
cor(y_num:acf_periodb.acf,y_num:acf_periodc.postMacf) 1.00 2574 3352
Population-Level Effects:
Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
Intercept -6.14 0.05 -6.24 -6.04 1.00 1003 1885
y_num -0.02 0.01 -0.03 -0.01 1.00 3625 3364
acf_periodb.acf 0.03 1.01 -1.99 2.02 1.00 3980 2998
acf_periodc.postMacf 0.05 0.10 -0.15 0.25 1.00 4687 3536
y_num:acf_periodb.acf 0.08 0.13 -0.17 0.33 1.00 3970 3191
y_num:acf_periodc.postMacf -0.05 0.01 -0.07 -0.03 1.00 4692 3159
Family Specific Parameters:
Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
shape 91.79 20.34 60.88 140.54 1.00 2971 2721
Draws were sampled using sampling(NUTS). For each parameter, Bulk_ESS
and Tail_ESS are effective sample size measures, and Rhat is the potential
scale reduction factor on split chains (at convergence, Rhat = 1).
plot(m_pulmonary)
pp_check(m_pulmonary, type='ecdf_overlay')
Using 10 posterior draws for ppc type 'ecdf_overlay' by default.
Summarise the posterior in graphical form
prior_summary(m_pulmonary)
prior class coef group resp dpar nlpar lb ub
normal(0, 1) b
normal(0, 1) b acf_periodb.acf
normal(0, 1) b acf_periodc.postMacf
normal(0, 1) b y_num
normal(0, 1) b y_num:acf_periodb.acf
normal(0, 1) b y_num:acf_periodc.postMacf
normal(0, 1) Intercept
lkj_corr_cholesky(4) L
lkj_corr_cholesky(4) L ward
exponential(1) sd 0
exponential(1) sd ward 0
exponential(1) sd acf_periodb.acf ward 0
exponential(1) sd acf_periodc.postMacf ward 0
exponential(1) sd Intercept ward 0
exponential(1) sd y_num ward 0
exponential(1) sd y_num:acf_periodb.acf ward 0
exponential(1) sd y_num:acf_periodc.postMacf ward 0
gamma(0.01, 0.01) shape 0
source
user
(vectorized)
(vectorized)
(vectorized)
(vectorized)
(vectorized)
user
user
(vectorized)
user
(vectorized)
(vectorized)
(vectorized)
(vectorized)
(vectorized)
(vectorized)
(vectorized)
user
f1b <- plot_counterfactual(model_data = overall_scaffold, model = m_pulmonary,
population_denominator = population_without_inst_ship, outcome = inc_100k, grouping_var=NULL,
re_formula = NA)
f1b
Make this into a figure combined with the map of empirical data
f1a <- st_as_sf(left_join(ward_inc, glasgow_wards_1951)) %>%
filter(tb_type=="Pulmonary") %>%
ggplot() +
geom_sf(aes(fill=inc_100k)) +
facet_wrap(year~., ncol = 7) +
scale_fill_viridis_c(name="Case notification rate (per 100,000)",
option = "A") +
theme_ggdist() +
theme(panel.border = element_rect(colour = "grey78", fill=NA),
legend.position = "top",
legend.key.width = unit(2, "cm"),
legend.title.align = 0.5,
text = element_text(size=8),
axis.text.x = element_text(angle=45, size=6, hjust=1),
axis.text.y = element_text(size=6)) +
guides(fill=guide_colorbar(title.position = "top"))
Joining with `by = join_by(division, ward, ward_number)`
(f1a / f1b) + plot_annotation(tag_levels = "A")
ggsave(here("figures/f1.png"))
Saving 7.29 x 4.51 in image
Summary of change in notifications numerically
overall_change <- summarise_change(model_data=overall_scaffold, model=m_pulmonary,
population_denominator=population_without_inst_ship, grouping_var=NULL, re_formula = NA)
`summarise()` has grouped output by '.draw'. You can override using the `.groups` argument.
#want to keep the summary estimates here
tokeep <- c("peak_summary", "level_summary", "slope_summary")
#summary measures in a table
overall_change %>%
keep((names(.) %in% tokeep)) %>%
bind_rows() %>%
mutate(across(c(estimate:.upper), number, accuracy=0.01)) %>%
select(measure, everything()) %>%
datatable()
Warning: There was 1 warning in `mutate()`.
ℹ In argument: `across(c(estimate:.upper), number, accuracy = 0.01)`.
Caused by warning:
! The `...` argument of `across()` is deprecated as of dplyr 1.1.0.
Supply arguments directly to `.fns` through an anonymous function instead.
# Previously
across(a:b, mean, na.rm = TRUE)
# Now
across(a:b, \(x) mean(x, na.rm = TRUE))
This warning is displayed once every 8 hours.
Call `lifecycle::last_lifecycle_warnings()` to see where this warning was generated.
NA
Numbers of pulmonary TB cases averted compared to counterfactual per year.
overall_pulmonary_counterf <- calculate_counterfactual(model_data = overall_scaffold, model=m_pulmonary, population_denominator = population_without_inst_ship)
Joining with `by = join_by(year, population_without_inst_ship, .draw)`Joining with `by = join_by(.draw)`
overall_pulmonary_counterf$counter_post %>%
mutate(across(c(cases_averted:cases_averted.upper, diff_inc100k:diff_inc100k.upper), number_format(accuracy = 0.1, big.mark = ","))) %>%
mutate(across(c(rr_inc100k:rr_inc100k.upper), number_format(accuracy = 0.01))) %>%
mutate(across(c(pct_change:pct_change.upper), percent, accuracy=0.1)) %>%
datatable()
NA
NA
Total pulmonary TB cases averted between 1958 and 1963
overall_pulmonary_counterf$counter_post_overall %>%
mutate(across(c(cases_averted:cases_averted.upper), number_format(accuracy = 0.1, big.mark = ","))) %>%
mutate(across(c(pct_change:pct_change.upper), percent, accuracy=0.1)) %>%
datatable()
NA
NA
What are the correlations between peak, level, and slope?
#RR.peak histogram
a <- overall_change$peak_draws %>%
ggplot() +
geom_histogram(aes(x=estimate), fill="darkblue", colour="darkblue", alpha=0.3)+
scale_fill_gradient(high="lightblue1",low="darkblue") +
theme_ggdist() +
theme(legend.position = "none",
panel.border = element_rect(colour = "grey78", fill=NA)) +
labs(x="RR.peak",
y="")
#RR. level histogram
b <- overall_change$level_draws %>%
ggplot() +
geom_histogram(aes(x=estimate), fill="darkblue", colour="darkblue", alpha=0.3)+
scale_fill_gradient(high="lightblue1",low="darkblue") +
theme_ggdist() +
theme(legend.position = "none",
panel.border = element_rect(colour = "grey78", fill=NA)) +
labs(x="RR.level",
y="")
#RR.slope histogram
c <- overall_change$slope_draws %>%
ggplot() +
geom_histogram(aes(x=estimate), fill="darkblue", colour="darkblue", alpha=0.3)+
scale_fill_gradient(high="lightblue1",low="darkblue") +
#scale_x_continuous(limits = c(0, 6)) +
theme_ggdist() +
theme(legend.position = "none",
panel.border = element_rect(colour = "grey78", fill=NA)) +
labs(x="RR.slope (logit scale)",
y="")
#Correlation between RR.peak and RR.level
cor_rr_peak_rr_level <- round(cor(pluck(overall_change$peak_draws$estimate), pluck(overall_change$level_draws$estimate)), digits = 2)
#Correlation between RR.peak and RR.slope
cor_rr_peak_rr_slope <- round(cor(pluck(overall_change$peak_draws$estimate), pluck(overall_change$slope_draws$estimate)), digits = 2)
#Correlation between RR.level and RR.slope
cor_rr_level_rr_slope <- round(cor(pluck(overall_change$level_draws$estimate), pluck(overall_change$slope_draws$estimate)), digits = 2)
#plot of correlation between RR.peak and RR.level
d <- bind_cols(RR.peak=pluck(overall_change$peak_draws$estimate),
RR.level =pluck(overall_change$level_draws$estimate)) %>%
ggplot(aes(y=RR.peak, x = RR.level)) +
geom_hex() +
geom_smooth(se=FALSE, colour="firebrick", method = "lm") +
geom_text(aes(y=2.2, x=0.58, label=cor_rr_peak_rr_level), colour="firebrick") +
scale_fill_gradient(high="lightblue1",low="darkblue") +
theme_ggdist() +
theme(legend.position = "none",
panel.border = element_rect(colour = "grey78", fill=NA))
#plot of correlation between RR.peak and RR.slope
e <- bind_cols(RR.peak=pluck(overall_change$peak_draws$estimate),
RR.slope =pluck(overall_change$slope_draws$estimate)) %>%
ggplot(aes(y=RR.peak, x = RR.slope)) +
geom_hex() +
geom_smooth(se=FALSE, colour="firebrick", method = "lm") +
geom_text(aes(y=2.1, x=0.5, label=cor_rr_peak_rr_slope), colour="firebrick") +
#scale_x_continuous(limits = c(0, 6)) +
scale_fill_gradient(high="lightblue1",low="darkblue") +
theme_ggdist() +
theme(legend.position = "none",
panel.border = element_rect(colour = "grey78", fill=NA))
#plot of correlation between RR.level and RR.slope
f <- bind_cols(RR.level=pluck(overall_change$level_draws$estimate),
RR.slope =pluck(overall_change$slope_draws$estimate)) %>%
ggplot(aes(y=RR.level, x = RR.slope)) +
geom_hex() +
geom_smooth(se=FALSE, colour="firebrick", method = "lm") +
geom_text(aes(y=0.75, x=0.5, label=cor_rr_level_rr_slope), colour="firebrick") +
#scale_x_continuous(limits = c(0, 6)) +
scale_fill_gradient(high="lightblue1",low="darkblue") +
theme_ggdist() +
theme(legend.position = "none",
panel.border = element_rect(colour = "grey78", fill=NA))
(plot_spacer() + plot_spacer() + c) /
(plot_spacer() + b + f) /
(a + d + e)
ℹ The package "hexbin" is required for `stat_binhex()`
✖ Would you like to install it?
1: Yes
2: No
1
trying URL 'https://cran.rstudio.com/bin/macosx/big-sur-arm64/contrib/4.3/hexbin_1.28.3.tgz'
Content type 'application/x-gzip' length 1609993 bytes (1.5 MB)
==================================================
downloaded 1.5 MB
The downloaded binary packages are in
/var/folders/n2/yxj0lwh51wq545d2xfnhh7740000gn/T//RtmpGldgYw/downloaded_packages
ggsave(here("figures/pulmonary_cors.pdf"), width=8, height=8)
NA
NA
NA
Plot the counterfactual at ward level
plot_counterfactual(model_data = mdata, model=m_pulmonary, outcome = inc_100k, population_denominator = population_without_inst_ship,
grouping_var = ward, ward, re_formula= ~(1 + y_num*acf_period | ward))
ggsave(here("figures/s3.png"), width=12, height=12)
Summary of change in notifications at ward level
ward_change <- summarise_change(model_data=mdata, model=m_pulmonary,
population_denominator=population_without_inst_ship, grouping_var=ward,
re_formula = ~(1 + y_num*acf_period | ward))
`summarise()` has grouped output by '.draw'. You can override using the `.groups` argument.`summarise()` has grouped output by '.draw'. You can override using the `.groups` argument.`summarise()` has grouped output by '.draw', 'acf_period'. You can override using the `.groups` argument.`summarise()` has grouped output by '.draw'. You can override using the `.groups` argument.
#want to keep the summary estimates here
tokeep <- c("peak_summary", "level_summary", "slope_summary")
#summary measures in a table
ward_change %>%
keep((names(.) %in% tokeep)) %>%
bind_rows() %>%
mutate(across(c(estimate:.upper), number, accuracy=0.01)) %>%
select(measure, everything()) %>%
datatable()
#plot these in a figure
ward_effects <- ward_change %>%
keep((names(.) %in% tokeep)) %>%
bind_rows() %>%
bind_rows(overall_change$peak_summary) %>%
bind_rows(overall_change$level_summary) %>%
bind_rows(overall_change$slope_summary) %>%
mutate_at(.vars = vars(estimate:.upper),
.funs = funs(as.numeric)) %>%
select(measure, everything()) %>%
mutate(estimate = as.double(estimate)) %>%
full_join(glasgow_wards_1951) %>%
mutate(ward2 = paste0(ward_number, ". ", ward)) %>%
mutate(ward2 = case_when(is.na(ward) ~ "Overall",
TRUE ~ ward2)) %>%
st_as_sf()
Warning: `funs()` was deprecated in dplyr 0.8.0.
Please use a list of either functions or lambdas:
# Simple named list:
list(mean = mean, median = median)
# Auto named with `tibble::lst()`:
tibble::lst(mean, median)
# Using lambdas
list(~ mean(., trim = .2), ~ median(., na.rm = TRUE))Joining with `by = join_by(ward)`
#function for plotting choropleth maps
plot_ward_effect <- function(data, measure){
{{data}} %>%
filter(measure == {{measure}}) %>%
ggplot() +
geom_sf(aes(fill=estimate)) +
geom_sf_label(aes(label = ward_number), size=3, fill=NA, label.size = NA, colour="black") +
scale_fill_gradient(high="lightblue1",low="darkblue", name="") +
theme_ggdist() +
theme(panel.border = element_rect(colour = "grey78", fill=NA),
axis.text.x=element_text(angle=45, hjust=1)) +
labs(x="", y="")
}
#function for plotting catapiller plots
plot_ward_cat <- function(data, measure, scale){
pal <- colorRampPalette(c('darkblue','lightblue'))
{{data}} %>%
filter(measure=={{measure}}) %>%
mutate(my_palette = case_when(ward2=="Overall" ~ "#C60C30",
TRUE ~ pal(36)[as.numeric(cut(.$estimate,breaks = 36))])) %>%
ggplot() +
geom_pointrange(aes(y=estimate, ymin=.lower, ymax=.upper,
x=fct_reorder(ward2, estimate), colour=my_palette)) +
coord_flip() +
scale_colour_identity(name="") +
scale_y_continuous() +
theme_ggdist() +
theme(panel.border = element_rect(colour = "grey78", fill=NA)) +
labs(x = "",
y = "Relative rate (95% CrI)")
}
ward_peak_i <- plot_ward_effect(data = ward_effects, measure = "RR.peak")
ward_level_i <- plot_ward_effect(data = ward_effects, measure = "RR.level")
ward_slope_i <- plot_ward_effect(data = ward_effects, measure = "RR.slope")
ward_peak_ii <- plot_ward_cat(data = ward_effects, measure = "RR.peak")
ward_level_ii <- plot_ward_cat(data = ward_effects, measure = "RR.level")
ward_slope_ii <- plot_ward_cat(data = ward_effects, measure = "RR.slope")
s4 <- (ward_peak_i + ward_level_i + ward_slope_i) /
(ward_peak_ii + ward_level_ii + ward_slope_ii)
s4[[1]] <- s4[[1]] + plot_layout(tag_level = 'new')
s4[[2]] <- s4[[2]] + plot_layout(tag_level = 'new')
s4 + plot_annotation(tag_levels = c('A', '1'))
ggsave(here("figures/s4.png"), width = 16, height=10)
Calculate the counterfactual per ward
ward_pulmonary_counterf <- calculate_counterfactual(model_data = mdata, model=m_pulmonary,
population_denominator = population_without_inst_ship,
grouping_var = ward, re_formula=~(1 + y_num*acf_period | ward))
`summarise()` has grouped output by '.draw'. You can override using the `.groups` argument.`summarise()` has grouped output by '.draw'. You can override using the `.groups` argument.Joining with `by = join_by(year, population_without_inst_ship, .draw, ward)`Joining with `by = join_by(.draw, ward)`
ward_pulmonary_counterf$counter_post %>%
mutate(across(c(cases_averted:cases_averted.upper, diff_inc100k:diff_inc100k.upper), number_format(accuracy = 0.1, big.mark = ","))) %>%
mutate(across(c(rr_inc100k:rr_inc100k.upper), number_format(accuracy = 0.01))) %>%
mutate(across(c(pct_change:pct_change.upper), percent, accuracy=0.1)) %>%
datatable()
NA
NA
Overall counterfactual per ward
ward_pulmonary_counterf$counter_post_overall %>%
mutate(across(c(cases_averted:cases_averted.upper), number_format(accuracy = 0.1, big.mark = ","))) %>%
mutate(across(c(pct_change:pct_change.upper), percent, accuracy=0.1)) %>%
datatable()
NA
Now we will model the extra-pulmonary TB notification rate. Struggling a bit with negative binomial model, so revert to Poisson.
m_extrapulmonary <- brm(
cases ~ 1 + y_num*acf_period + (1 + y_num*acf_period | ward) + offset(log(population_without_inst_ship)),
data = mdata_extrapulmonary,
family = negbinomial(),
seed = 1234,
chains = 4, cores = 4,
prior = prior(normal(0,1000), class = Intercept) +
prior(gamma(0.01, 0.01), class = shape) +
prior(normal(0, 1), class = b) +
prior(exponential(1), class=sd) +
prior(lkj(2), class=cor))
Compiling Stan program...
Start sampling
starting worker pid=10277 on localhost:11273 at 19:34:23.655
starting worker pid=10292 on localhost:11273 at 19:34:23.744
starting worker pid=10306 on localhost:11273 at 19:34:23.840
starting worker pid=10319 on localhost:11273 at 19:34:23.932
SAMPLING FOR MODEL 'anon_model' NOW (CHAIN 1).
Chain 1:
Chain 1: Gradient evaluation took 0.000345 seconds
Chain 1: 1000 transitions using 10 leapfrog steps per transition would take 3.45 seconds.
Chain 1: Adjust your expectations accordingly!
Chain 1:
Chain 1:
Chain 1: Iteration: 1 / 2000 [ 0%] (Warmup)
SAMPLING FOR MODEL 'anon_model' NOW (CHAIN 2).
Chain 2:
Chain 2: Gradient evaluation took 0.000157 seconds
Chain 2: 1000 transitions using 10 leapfrog steps per transition would take 1.57 seconds.
Chain 2: Adjust your expectations accordingly!
Chain 2:
Chain 2:
Chain 2: Iteration: 1 / 2000 [ 0%] (Warmup)
SAMPLING FOR MODEL 'anon_model' NOW (CHAIN 3).
Chain 3:
Chain 3: Gradient evaluation took 0.000151 seconds
Chain 3: 1000 transitions using 10 leapfrog steps per transition would take 1.51 seconds.
Chain 3: Adjust your expectations accordingly!
Chain 3:
Chain 3:
Chain 3: Iteration: 1 / 2000 [ 0%] (Warmup)
SAMPLING FOR MODEL 'anon_model' NOW (CHAIN 4).
Chain 4:
Chain 4: Gradient evaluation took 0.000159 seconds
Chain 4: 1000 transitions using 10 leapfrog steps per transition would take 1.59 seconds.
Chain 4: Adjust your expectations accordingly!
Chain 4:
Chain 4:
Chain 4: Iteration: 1 / 2000 [ 0%] (Warmup)
Chain 2: Iteration: 200 / 2000 [ 10%] (Warmup)
Chain 1: Iteration: 200 / 2000 [ 10%] (Warmup)
Chain 3: Iteration: 200 / 2000 [ 10%] (Warmup)
Chain 4: Iteration: 200 / 2000 [ 10%] (Warmup)
Chain 2: Iteration: 400 / 2000 [ 20%] (Warmup)
Chain 1: Iteration: 400 / 2000 [ 20%] (Warmup)
Chain 3: Iteration: 400 / 2000 [ 20%] (Warmup)
Chain 4: Iteration: 400 / 2000 [ 20%] (Warmup)
Chain 2: Iteration: 600 / 2000 [ 30%] (Warmup)
Chain 1: Iteration: 600 / 2000 [ 30%] (Warmup)
Chain 2: Iteration: 800 / 2000 [ 40%] (Warmup)
Chain 3: Iteration: 600 / 2000 [ 30%] (Warmup)
Chain 4: Iteration: 600 / 2000 [ 30%] (Warmup)
Chain 2: Iteration: 1000 / 2000 [ 50%] (Warmup)
Chain 2: Iteration: 1001 / 2000 [ 50%] (Sampling)
Chain 1: Iteration: 800 / 2000 [ 40%] (Warmup)
Chain 3: Iteration: 800 / 2000 [ 40%] (Warmup)
Chain 4: Iteration: 800 / 2000 [ 40%] (Warmup)
Chain 2: Iteration: 1200 / 2000 [ 60%] (Sampling)
Chain 3: Iteration: 1000 / 2000 [ 50%] (Warmup)
Chain 3: Iteration: 1001 / 2000 [ 50%] (Sampling)
Chain 1: Iteration: 1000 / 2000 [ 50%] (Warmup)
Chain 1: Iteration: 1001 / 2000 [ 50%] (Sampling)
Chain 4: Iteration: 1000 / 2000 [ 50%] (Warmup)
Chain 4: Iteration: 1001 / 2000 [ 50%] (Sampling)
Chain 2: Iteration: 1400 / 2000 [ 70%] (Sampling)
Chain 3: Iteration: 1200 / 2000 [ 60%] (Sampling)
Chain 1: Iteration: 1200 / 2000 [ 60%] (Sampling)
Chain 4: Iteration: 1200 / 2000 [ 60%] (Sampling)
Chain 2: Iteration: 1600 / 2000 [ 80%] (Sampling)
Chain 3: Iteration: 1400 / 2000 [ 70%] (Sampling)
Chain 2: Iteration: 1800 / 2000 [ 90%] (Sampling)
Chain 3: Iteration: 1600 / 2000 [ 80%] (Sampling)
Chain 1: Iteration: 1400 / 2000 [ 70%] (Sampling)
Chain 4: Iteration: 1400 / 2000 [ 70%] (Sampling)
Chain 2: Iteration: 2000 / 2000 [100%] (Sampling)
Chain 2:
Chain 2: Elapsed Time: 8.244 seconds (Warm-up)
Chain 2: 4.206 seconds (Sampling)
Chain 2: 12.45 seconds (Total)
Chain 2:
Chain 3: Iteration: 1800 / 2000 [ 90%] (Sampling)
Chain 1: Iteration: 1600 / 2000 [ 80%] (Sampling)
Chain 4: Iteration: 1600 / 2000 [ 80%] (Sampling)
Chain 3: Iteration: 2000 / 2000 [100%] (Sampling)
Chain 3:
Chain 3: Elapsed Time: 9.136 seconds (Warm-up)
Chain 3: 4.067 seconds (Sampling)
Chain 3: 13.203 seconds (Total)
Chain 3:
Chain 4: Iteration: 1800 / 2000 [ 90%] (Sampling)
Chain 1: Iteration: 1800 / 2000 [ 90%] (Sampling)
Chain 4: Iteration: 2000 / 2000 [100%] (Sampling)
Chain 4:
Chain 4: Elapsed Time: 9.422 seconds (Warm-up)
Chain 4: 5.736 seconds (Sampling)
Chain 4: 15.158 seconds (Total)
Chain 4:
Chain 1: Iteration: 2000 / 2000 [100%] (Sampling)
Chain 1:
Chain 1: Elapsed Time: 9.347 seconds (Warm-up)
Chain 1: 6.205 seconds (Sampling)
Chain 1: 15.552 seconds (Total)
Chain 1:
summary(m_extrapulmonary)
Family: negbinomial
Links: mu = log; shape = identity
Formula: cases ~ 1 + y_num * acf_period + (1 + y_num * acf_period | ward) + offset(log(population_without_inst_ship))
Data: mdata_extrapulmonary (Number of observations: 444)
Draws: 4 chains, each with iter = 2000; warmup = 1000; thin = 1;
total post-warmup draws = 4000
Group-Level Effects:
~ward (Number of levels: 37)
Estimate Est.Error l-95% CI u-95% CI
sd(Intercept) 0.33 0.06 0.23 0.45
sd(y_num) 0.02 0.01 0.00 0.05
sd(acf_periodb.acf) 0.11 0.09 0.00 0.33
sd(acf_periodc.postMacf) 0.13 0.10 0.00 0.36
sd(y_num:acf_periodb.acf) 0.01 0.01 0.00 0.04
sd(y_num:acf_periodc.postMacf) 0.02 0.01 0.00 0.04
cor(Intercept,y_num) -0.11 0.31 -0.65 0.52
cor(Intercept,acf_periodb.acf) -0.01 0.33 -0.64 0.59
cor(y_num,acf_periodb.acf) -0.02 0.33 -0.63 0.62
cor(Intercept,acf_periodc.postMacf) -0.07 0.33 -0.66 0.59
cor(y_num,acf_periodc.postMacf) -0.06 0.33 -0.66 0.58
cor(acf_periodb.acf,acf_periodc.postMacf) 0.01 0.33 -0.62 0.62
cor(Intercept,y_num:acf_periodb.acf) -0.01 0.33 -0.64 0.62
cor(y_num,y_num:acf_periodb.acf) -0.01 0.33 -0.63 0.60
cor(acf_periodb.acf,y_num:acf_periodb.acf) -0.08 0.34 -0.71 0.58
cor(acf_periodc.postMacf,y_num:acf_periodb.acf) 0.01 0.32 -0.62 0.63
cor(Intercept,y_num:acf_periodc.postMacf) -0.15 0.32 -0.69 0.51
cor(y_num,y_num:acf_periodc.postMacf) -0.06 0.33 -0.66 0.59
cor(acf_periodb.acf,y_num:acf_periodc.postMacf) 0.03 0.33 -0.62 0.65
cor(acf_periodc.postMacf,y_num:acf_periodc.postMacf) -0.10 0.35 -0.73 0.58
cor(y_num:acf_periodb.acf,y_num:acf_periodc.postMacf) 0.02 0.33 -0.62 0.64
Rhat Bulk_ESS Tail_ESS
sd(Intercept) 1.00 1525 2401
sd(y_num) 1.01 503 758
sd(acf_periodb.acf) 1.00 1672 1222
sd(acf_periodc.postMacf) 1.00 1456 1707
sd(y_num:acf_periodb.acf) 1.00 1905 1936
sd(y_num:acf_periodc.postMacf) 1.00 913 1038
cor(Intercept,y_num) 1.00 2043 2575
cor(Intercept,acf_periodb.acf) 1.00 4028 2840
cor(y_num,acf_periodb.acf) 1.00 4054 3108
cor(Intercept,acf_periodc.postMacf) 1.00 4348 2941
cor(y_num,acf_periodc.postMacf) 1.00 3233 2627
cor(acf_periodb.acf,acf_periodc.postMacf) 1.00 2885 3130
cor(Intercept,y_num:acf_periodb.acf) 1.00 4039 2716
cor(y_num,y_num:acf_periodb.acf) 1.00 3474 2967
cor(acf_periodb.acf,y_num:acf_periodb.acf) 1.00 3227 3122
cor(acf_periodc.postMacf,y_num:acf_periodb.acf) 1.00 3346 3296
cor(Intercept,y_num:acf_periodc.postMacf) 1.00 3021 2605
cor(y_num,y_num:acf_periodc.postMacf) 1.00 2757 2667
cor(acf_periodb.acf,y_num:acf_periodc.postMacf) 1.00 2628 3042
cor(acf_periodc.postMacf,y_num:acf_periodc.postMacf) 1.00 1975 3043
cor(y_num:acf_periodb.acf,y_num:acf_periodc.postMacf) 1.00 2328 2310
Population-Level Effects:
Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
Intercept -7.93 0.07 -8.07 -7.78 1.00 1642 2527
y_num -0.09 0.01 -0.12 -0.07 1.00 3832 3404
acf_periodb.acf -0.03 0.98 -1.93 1.93 1.00 2760 2715
acf_periodc.postMacf -0.35 0.39 -1.11 0.41 1.00 2294 2381
y_num:acf_periodb.acf -0.01 0.12 -0.26 0.23 1.00 2723 2738
y_num:acf_periodc.postMacf 0.02 0.04 -0.05 0.10 1.00 2115 2245
Family Specific Parameters:
Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
shape 92.27 66.10 26.85 269.28 1.00 3436 2750
Draws were sampled using sampling(NUTS). For each parameter, Bulk_ESS
and Tail_ESS are effective sample size measures, and Rhat is the potential
scale reduction factor on split chains (at convergence, Rhat = 1).
plot(m_extrapulmonary)
pp_check(m_extrapulmonary, type='ecdf_overlay')
Using 10 posterior draws for ppc type 'ecdf_overlay' by default.
Summarise in plot
plot_counterfactual(model_data = overall_scaffold %>% filter(year<=1961), model=m_extrapulmonary,
population_denominator = population_without_inst_ship, outcome=inc_100k_extrapulmonary, re_formula = NA)
ggsave(here("figures/s6.png"), width=10)
Saving 10 x 4.51 in image
Summarise numerically.
overall_change_extrapulmonary <- summarise_change(model_data=overall_scaffold, model=m_extrapulmonary,
population_denominator=population_without_inst_ship, grouping_var=NULL, re_formula = NA)
`summarise()` has grouped output by '.draw'. You can override using the `.groups` argument.
#want to keep the summary estimates here
tokeep <- c("peak_summary", "level_summary", "slope_summary")
#summary measures in a table
overall_change_extrapulmonary %>%
keep(names(.) %in% tokeep) %>%
bind_rows() %>%
mutate(across(c(estimate:.upper), number, accuracy=0.01)) %>%
select(measure, everything()) %>%
datatable()
NA
Numbers of extra-pulmonary TB cases averted overall.
overall_ep_counterf <- calculate_counterfactual(model_data = mdata_extrapulmonary, model=m_extrapulmonary,
population_denominator = population_without_inst_ship)
Joining with `by = join_by(year, population_without_inst_ship, .draw)`Joining with `by = join_by(.draw)`
overall_ep_counterf$counter_post %>%
mutate(across(c(cases_averted:cases_averted.upper, diff_inc100k:diff_inc100k.upper), number_format(accuracy = 0.1, big.mark = ","))) %>%
mutate(across(c(rr_inc100k:rr_inc100k.upper), number_format(accuracy = 0.01))) %>%
mutate(across(c(pct_change:pct_change.upper), percent, accuracy=0.1)) %>%
datatable()
NA
Total extrapulmonary TB cases averted between 1958 and 1963
overall_ep_counterf$counter_post_overall %>%
mutate(across(c(cases_averted:cases_averted.upper), number_format(accuracy = 0.1, big.mark = ","))) %>%
mutate(across(c(pct_change:pct_change.upper), percent, accuracy=0.1)) %>%
datatable()
NA
NA
Ward-level extra-pulmonary estimates in graphical form.
plot_counterfactual(model_data = mdata_extrapulmonary, model=m_extrapulmonary, outcome = inc_100k,
population_denominator = population_without_inst_ship, grouping_var = ward,re_formula =~(y_num*acf_period | ward),
ward)
ggsave(here("figures/s4.png"), width=10, height=12)
Numerical summary.
ward_change_extrapulmonary <- summarise_change(model_data = mdata_extrapulmonary, model = m_extrapulmonary,
population_denominator = population_without_inst_ship, grouping_var=ward,
re_formula = ~(y_num*acf_period | ward))
`summarise()` has grouped output by '.draw'. You can override using the `.groups` argument.`summarise()` has grouped output by '.draw'. You can override using the `.groups` argument.`summarise()` has grouped output by '.draw', 'acf_period'. You can override using the `.groups` argument.`summarise()` has grouped output by '.draw'. You can override using the `.groups` argument.
#want to keep the summary estimates here
tokeep <- c("peak_summary", "level_summary", "slope_summary")
#summary measures in a table
ward_change_extrapulmonary %>%
keep(names(.) %in% tokeep) %>%
bind_rows() %>%
mutate(across(c(estimate:.upper), number, accuracy=0.01)) %>%
select(measure, everything()) %>%
datatable()
NA
NA
NA
Fit the model
(Not rewritten the functions for this yet)
Summarise posterior
ggsave(here("figures/s7.png"), height=10)
Saving 7 x 10 in image
Calculate summary effects
level_draws_age_sex <- add_epred_draws(newdata = out_age_sex_2,
object = m_age_sex) %>%
arrange(y_num, .draw) %>%
group_by(.draw, age, sex) %>%
summarise(estimate = last(.epred)/first(.epred)) %>%
ungroup() %>%
mutate(measure = "RR.level")
`summarise()` has grouped output by '.draw', 'age'. You can override using the `.groups` argument.
Numerica sumamry of these summary results
As a figure
peak_g_age_sex <- peak_summary_age_sex %>%
mutate(sex = case_when(sex=="M" ~ "Male",
sex=="F" ~ "Female")) %>%
mutate(age = case_when(age=="00_05" ~ "0 to 5y",
age=="06_15" ~ "06 to 15y",
age=="16_25" ~ "16 to 25y",
age=="26_35" ~ "26 to 35y",
age=="36_45" ~ "36 to 45y",
age=="46_55" ~ "46 to 55y",
age=="56_65" ~ "56 to 65y",
age=="65+" ~ "65 & up y")) %>%
ggplot() +
geom_hline(aes(yintercept=1), linetype=2)+
geom_pointrange(aes(x=age, y=estimate, ymin=.lower, ymax=.upper, group=sex, colour=sex, shape=sex),
position = position_dodge(width = 0.5)) +
scale_colour_manual(values = c("#CD7AC5", "cadetblue3"), name="") +
scale_shape(name="") +
labs(x="",
y="Relative rate (95% UI)") +
theme_ggdist() +
theme(legend.position = "bottom",
panel.border = element_rect(colour = "grey78", fill=NA))
#level plot
level_g_age_sex <- level_summary_age_sex %>%
mutate(sex = case_when(sex=="M" ~ "Male",
sex=="F" ~ "Female")) %>%
mutate(age = case_when(age=="00_05" ~ "0 to 5y",
age=="06_15" ~ "06 to 15y",
age=="16_25" ~ "16 to 25y",
age=="26_35" ~ "26 to 35y",
age=="36_45" ~ "36 to 45y",
age=="46_55" ~ "46 to 55y",
age=="56_65" ~ "56 to 65y",
age=="65+" ~ "65 & up y")) %>%
ggplot() +
geom_hline(aes(yintercept=1), linetype=2)+
geom_pointrange(aes(x=age, y=estimate, ymin=.lower, ymax=.upper, group=sex, colour=sex, shape=sex),
position = position_dodge(width = 0.5)) +
scale_colour_manual(values = c("#CD7AC5", "cadetblue3"), name="") +
scale_shape(name="") +
labs(x="",
y="Relative rate (95% UI)") +
theme_ggdist() +
theme(legend.position = "bottom",
panel.border = element_rect(colour = "grey78", fill=NA))
#slope plot
slope_g_age_sex <- slope_summary_age_sex %>%
mutate(sex = case_when(sex=="M" ~ "Male",
sex=="F" ~ "Female")) %>%
mutate(age = case_when(age=="00_05" ~ "0 to 5y",
age=="06_15" ~ "06 to 15y",
age=="16_25" ~ "16 to 25y",
age=="26_35" ~ "26 to 35y",
age=="36_45" ~ "36 to 45y",
age=="46_55" ~ "46 to 55y",
age=="56_65" ~ "56 to 65y",
age=="65+" ~ "65 & up y")) %>%
ggplot() +
geom_hline(aes(yintercept=1), linetype=2)+
geom_pointrange(aes(x=age, y=estimate, ymin=.lower, ymax=.upper, group=sex, colour=sex, shape=sex),
position = position_dodge(width = 0.5)) +
scale_colour_manual(values = c("#CD7AC5", "cadetblue3"), name="") +
scale_shape(name="") +
labs(x="",
y="Relative rate (95% UI)") +
theme_ggdist() +
theme(legend.position = "bottom",
panel.border = element_rect(colour = "grey78", fill=NA))
counterfact_age_sex <-
add_epred_draws(object = m_age_sex,
newdata = mdata_age_sex %>%
select(year, year2, y_num, age, sex) %>%
mutate(acf_period = "a. pre-acf")) %>%
filter(year>1957) %>%
select(year, age, sex, .draw, .epred_counterf = .epred)
Adding missing grouping variables: `year2`, `y_num`, `acf_period`, `.row`
#Calcuate predicted number of cases per draw, then summarise.
post_change_age_sex <-
add_epred_draws(object = m_age_sex,
newdata = mdata_age_sex %>%
select(year, year2, y_num, age, sex, acf_period)) %>%
filter(year>1957) %>%
ungroup() %>%
select(year, age, sex, .draw, .epred)
#for the overall period
counterfact_overall_age_sex <-
add_epred_draws(object = m_age_sex,
newdata = mdata_age_sex %>%
select(year, year2, y_num, age, sex) %>%
mutate(acf_period = "a. pre-acf")) %>%
filter(year>1957) %>%
select(age, sex, .draw, .epred) %>%
group_by(age, sex, .draw) %>%
summarise(.epred_counterf = sum(.epred)) %>%
mutate(year = "Overall (1958-1963)")
Adding missing grouping variables: `year`, `year2`, `y_num`, `acf_period`, `.row``summarise()` has grouped output by 'age', 'sex'. You can override using the `.groups` argument.
#Calcuate incidence per draw, then summarise.
post_change_overall_age_sex <-
add_epred_draws(object = m_age_sex,
newdata = mdata_age_sex %>%
select(year, year2, y_num, age, sex, acf_period)) %>%
filter(year>1957) %>%
select(age, sex, .draw, .epred) %>%
group_by(.draw, age, sex) %>%
summarise(.epred = sum(.epred))
Adding missing grouping variables: `year`, `year2`, `y_num`, `acf_period`, `.row``summarise()` has grouped output by '.draw', 'age'. You can override using the `.groups` argument.
counter_post_overall_age_sex <-
left_join(counterfact_overall_age_sex, post_change_overall_age_sex) %>%
mutate(cases_averted = .epred_counterf-.epred,
pct_change = (.epred - .epred_counterf)/.epred_counterf) %>%
group_by(age, sex) %>%
mean_qi(cases_averted, pct_change) %>%
ungroup() %>%
mutate(year = "Overall (1958-1963)")
Joining with `by = join_by(age, sex, .draw)`
age_sex_txt <- counter_post_overall_age_sex %>%
mutate(across(c(cases_averted:cases_averted.upper), number_format(accuracy = 0.1, big.mark = ","))) %>%
mutate(across(c(pct_change:pct_change.upper), percent, accuracy=0.1)) %>%
transmute(year = as.character(year),
sex = sex,
age = age,
cases_averted = glue::glue("{cases_averted}\n({cases_averted.lower} to {cases_averted.upper})"),
pct_change = glue::glue("{pct_change}\n({pct_change.lower} to {pct_change.upper})"))
age_sex_txt %>% datatable()
NA
NA
Join together for Figure 3.